AIJ - 24

• (2025). Unsupervised sentence selection for creating a representative corpus in turkish: An active learning approach. AIJ, 348, 104422. (www)

  @article{ , abstract = {In this study, active learning methods adapted for sentence selection of Turkish sentences are evaluated through language learning with neural models. Turkish is an agglutinative language with a complex morphology, where the linguistic properties of words are encoded in suffixes. The active learning methods based on regression, clustering, language models, distance metrics, and neural networks are applied to unlabeled sentence selection. In this respect, a sentence corpus is selected from a larger corpus, with the same number of samples for each target word in intrinsic and extrinsic evaluation tasks. The selected sentences are used for the training of SkipGram, CBOW, and self-attention LSTM language models and extracted embeddings are evaluated by the semantic analogy, POS and sentiment analysis tasks. The evaluation scores of the models trained on the samples selected by the active learning method are compared. The results of the selected sentences based on language models indicate an improvement over random selection based on a static vocabulary. These results also show that the selection affects the quality of unsupervised word embedding extraction even if the target vocabulary is kept the same. Along with the accuracy, the time efficiency of the language models is shown to be better than other methods especially methods based on neural network models, and distance metrics.}, archive = {J_AIJ}, author = {Hayri Volkan Agun}, doi = {10.1016/j.artint.2025.104422}, journal = {Artificial Intelligence}, month = {11}, pages = {104422}, shortjournal = {Artif. Intell.}, title = {Unsupervised sentence selection for creating a representative corpus in turkish: An active learning approach}, volume = {348}, year = {2025}, }
• (2025). Learngene: Inheritable “genes” in intelligent agents. AIJ, 348, 104421. (www)

  @article{ , abstract = {Biological intelligence has driven significant progress in artificial intelligence (AI), but a critical gap remains: biological systems inherit innate abilities from genes, with brains initialized by blueprints refined over 3.5 billion years of evolution, while machines rely heavily on inefficient, data-driven learning from scratch. This gap arises from the lack of a genetic mechanism in machines to transfer and accumulate inheritable knowledge across generations. To bridge this gap, we propose learngenes, network fragments that act as inheritable “genes” for machines. Unlike conventional knowledge transfer methods, learngenes enable efficient and universal knowledge transfer by selectively encapsulating task-agnostic knowledge. To facilitate the transfer and accumulation of task-agnostic knowledge across generations, we introduce Genetic Reinforcement Learning (GRL), a framework that simulates the learning and evolution of organisms in intelligent agents following Lamarckian principles. Through GRL, we identify learngenes as network fragments within agents' policy networks, equipping newborn agents with innate abilities for rapid adaptation to novel tasks. We demonstrate the advantages of learngene-based knowledge transfer over evolution-based search and traditional pre-trained models, and show how learngenes evolve through the accumulation of task-agnostic knowledge. Overall, this work establishes a novel paradigm for knowledge transfer and model initialization in AI, offering new possibilities for more adaptive, efficient, and scalable learning systems.}, archive = {J_AIJ}, author = {Fu Feng and Jing Wang and Xu Yang and Xin Geng}, doi = {10.1016/j.artint.2025.104421}, journal = {Artificial Intelligence}, month = {11}, pages = {104421}, shortjournal = {Artif. Intell.}, title = {Learngene: Inheritable “genes” in intelligent agents}, volume = {348}, year = {2025}, }
• (2025). Bridging theory and practice in bidirectional heuristic search with front-to-end consistent heuristics. AIJ, 348, 104420. (www)

  @article{ , abstract = {Recent research on bidirectional heuristic search (BiHS) has been shaped by the must-expand pairs (MEP) theory, which identifies the pairs of nodes that must be expanded to ensure solution optimality. Another line of research has focused on algorithms utilizing lower bounds derived from consistent heuristics during the search. This paper bridges these two approaches, offering a unified framework that demonstrates how both existing and novel algorithms can be derived from MEP theory. We introduce an extended set of bounds, encompassing both previously known and newly formulated ones. Using these bounds, we develop a range of algorithms, each employing different criteria for termination, node selection, and search direction. Finally, we empirically evaluate how these bounds and algorithms impact search efficiency.}, archive = {J_AIJ}, author = {Lior Siag and Shahaf S. Shperberg}, doi = {10.1016/j.artint.2025.104420}, journal = {Artificial Intelligence}, month = {11}, pages = {104420}, shortjournal = {Artif. Intell.}, title = {Bridging theory and practice in bidirectional heuristic search with front-to-end consistent heuristics}, volume = {348}, year = {2025}, }
• (2025). Minimax off-policy evaluation and learning with subgaussian and differentiable importance weighting. AIJ, 348, 104419. (www)

  @article{ , abstract = {In this work, we study the statistical properties of the off-policy estimation problem, i.e., estimating expectations under a target policy using samples collected from a different policy. We begin by presenting a novel minimax concentration lower bound that highlights the fundamental limits of off-policy estimation. We then analyze two well-known importance weighting (IW) techniques: vanilla IW and self-normalized importance weighting (SN). For both methods, we derive concentration and anti-concentration results, showing that their concentration rates are provably suboptimal compared to our lower bound. Observing that this undesired behavior arises from the heavy-tailed nature of the IW and SN estimators, we propose a new class of parametric estimators based on a transformation using the power mean (PM), which is no longer heavy-tailed. We study the theoretical properties of the PM estimator in terms of bias and variance. We show that, with suitable (possibly data-driven) tuning of its parameters, the PM estimator satisfies two key properties under certain conditions: ( i ) it achieves a subgaussian concentration rate that matches our lower bound and ( ii ) it maintains differentiability with respect to the target policy. Finally, we validate our approach through numerical simulations on both synthetic datasets and contextual bandits, comparing it against standard off-policy evaluation and learning baselines. 1}, archive = {J_AIJ}, author = {Alberto Maria Metelli and Alessio Russo and Marcello Restelli}, doi = {10.1016/j.artint.2025.104419}, journal = {Artificial Intelligence}, month = {11}, pages = {104419}, shortjournal = {Artif. Intell.}, title = {Minimax off-policy evaluation and learning with subgaussian and differentiable importance weighting}, volume = {348}, year = {2025}, }
• (2025). On the disjunctive rational closure of a conditional knowledge base. AIJ, 348, 104418. (www)

  @article{ , abstract = {One of the most widely investigated decision problems in symbolic AI is that of which conditional sentences of the form “if α , then normally β ” should follow from a knowledge base containing this type of statements. Probably, the most notable approach to this problem is the rational closure construction put forward by Lehmann and Magidor in the'90s, which has been adapted to logical languages of various expressive powers since then. At the core of rational closure is the Rational Monotonicity property, which allows one to retain existing (defeasible) conclusions whenever new information cannot be negated by existing conclusions. As it turns out, Rational Monotonicity is not universally accepted, with many researchers advocating the investigation of weaker versions thereof leading to a larger class of consequence relations. A case in point is that of the Disjunctive Rationality property, which states that if one may draw a (defeasible) conclusion from a disjunction of premises, then one should be able to draw this conclusion from at least one of the premises taken alone. While there are convincing arguments that the rational closure forms the ‘simplest’ rational consequence relation extending a given set of conditionals, the question of what the simplest disjunctive consequence relation in this setting is has not been explored in depth. In this article, we do precisely that by motivating and proposing a concrete construction of the disjunctive rational closure of a conditional knowledge base, of which the properties and consequences of its adoption we also investigate in detail. (Previous versions of this work have been selected for presentation at the 18th International Workshop on Nonmonotonic Reasoning (NMR 2020) [1] and at the 35th AAAI Conference on Artificial Intelligence (AAAI 2021) [2] . The present submission extends and elaborates on both papers.)}, archive = {J_AIJ}, author = {Richard Booth and Ivan Varzinczak}, doi = {10.1016/j.artint.2025.104418}, journal = {Artificial Intelligence}, month = {11}, pages = {104418}, shortjournal = {Artif. Intell.}, title = {On the disjunctive rational closure of a conditional knowledge base}, volume = {348}, year = {2025}, }
• (2025). Rethinking visual prompt learning as masked visual token modeling. AIJ, 348, 104417. (www)

  @article{ , abstract = {Prompt learning has achieved great success in efficiently exploiting large-scale pre-trained models in natural language processing (NLP). It reformulates the downstream tasks as the generative pre-training ones to achieve consistency, thus improving the performance stably. However, when transferring it to the vision area, current visual prompt learning methods are almost designed on discriminative pre-trained models, and there is also a lack of careful design to unify the forms of pre-training and downstream tasks. To explore prompt learning on the generative pre-trained visual model, as well as keeping the task consistency, we propose Visual Prompt learning as masked visual Token Modeling (VPTM) to transform the downstream visual classification task into the pre-trained masked visual token prediction task. In addition, we develop the prototypical verbalizer for mapping the predicted visual token with implicit semantics to explicit downstream labels. To our best knowledge, VPTM is the first visual prompt method on the generative pre-trained visual model, which achieves consistency between pre-training and downstream visual classification by task reformulation. Experiments show that VPTM outperforms other visual prompt methods and achieves excellent efficiency. Moreover, the task consistency of VPTM contributes to the robustness against prompt location, prompt length and prototype dimension, and could be deployed uniformly.}, archive = {J_AIJ}, author = {Ning Liao and Bowen Shi and Xiaopeng Zhang and Min Cao and Junchi Yan and Qi Tian}, doi = {10.1016/j.artint.2025.104417}, journal = {Artificial Intelligence}, month = {11}, pages = {104417}, shortjournal = {Artif. Intell.}, title = {Rethinking visual prompt learning as masked visual token modeling}, volume = {348}, year = {2025}, }
• (2025). Planning for temporally extended goals in pure-past linear temporal logic. AIJ, 348, 104409. (www)

  @article{ , abstract = {We study planning for temporally extended goals expressed in Pure-Past Linear Temporal Logic ( ppltl ) in the context of deterministic (i.e., classical) and fully observable nondeterministic (FOND) domains. ppltl is the variant of Linear-time Temporal Logic on finite traces ( ltl f ) that refers to the past rather than the future. Although ppltl is as expressive as ltl f , we show that it is computationally much more effective for planning. In particular, we show that checking the validity of a plan for a ppltl formula is Markovian. This is achieved by introducing a linear number of additional propositional variables that capture the validity of the entire formula in a modular fashion. The solution encoding introduces only a linear number of new fluents proportional to the size of the ppltl goal and does not require any additional spurious action. We implement our solution technique in a system called Plan4Past , which can be used alongside state-of-the-art classical and FOND planners. Our empirical analysis demonstrates the practical effectiveness of Plan4Past in both classical and FOND problems, showing that the resulting planner performs overall better than other planning approaches for ltl f goals.}, archive = {J_AIJ}, author = {Luigi Bonassi and Giuseppe De Giacomo and Marco Favorito and Francesco Fuggitti and Alfonso Emilio Gerevini and Enrico Scala}, doi = {10.1016/j.artint.2025.104409}, journal = {Artificial Intelligence}, month = {11}, pages = {104409}, shortjournal = {Artif. Intell.}, title = {Planning for temporally extended goals in pure-past linear temporal logic}, volume = {348}, year = {2025}, }
• (2025). Incentives for responsiveness, instrumental control and impact. AIJ, 348, 104408. (www)

  @article{ , abstract = {We introduce three concepts that describe an agent's incentives: response incentives indicate which variables in the environment, such as sensitive demographic information, affect the decision under the optimal policy. Instrumental control incentives indicate whether an agent's policy is chosen to manipulate part of its environment, such as the preferences or instructions of a user. Impact incentives indicate which variables an agent will affect, intentionally or otherwise. For each concept, we establish sound and complete graphical criteria, and discuss general classes of techniques that may be used to produce incentives for safe and fair agent behaviour. Finally, we outline how these notions may be generalised to multi-decision settings. This journal paper extends our conference publication “Agent Incentives: A Causal Perspective”: the material on response incentives and instrumental control incentives is updated, while the work on impact incentives and multi-decision settings is entirely new.}, archive = {J_AIJ}, author = {Ryan Carey and Eric Langlois and Chris van Merwijk and Shane Legg and Tom Everitt}, doi = {10.1016/j.artint.2025.104408}, journal = {Artificial Intelligence}, month = {11}, pages = {104408}, shortjournal = {Artif. Intell.}, title = {Incentives for responsiveness, instrumental control and impact}, volume = {348}, year = {2025}, }
• (2025). Abstracting situation calculus action theories. AIJ, 348, 104407. (www)

  @article{ , abstract = {We develop a general framework for agent abstraction based on the situation calculus and the ConGolog agent programming language. We assume that we have a high-level specification and a low-level specification of the agent, both represented as basic action theories. A refinement mapping specifies how each high-level action is implemented by a low-level ConGolog program and how each high-level fluent can be translated into a low-level formula. We define a notion of sound abstraction between such action theories in terms of the existence of a suitable bisimulation between their respective models. Sound abstractions have many useful properties that ensure that we can reason about the agent's actions (e.g., executability, projection, and planning) at the abstract level, and refine and concretely execute them at the low level. We also characterize the notion of complete abstraction where all actions (including exogenous ones) that the high level thinks can happen can in fact occur at the low level. To facilitate verifying that one has a sound/complete abstraction relative to a mapping, we provide a set of necessary and sufficient conditions. Finally, we identify a set of basic action theory constraints that ensure that for any low-level action sequence, there is a unique high-level action sequence that it refines. This allows us to track/monitor what the low-level agent is doing and describe it in abstract terms (i.e., provide high-level explanations, for instance, to a client or manager).}, archive = {J_AIJ}, author = {Bita Banihashemi and Giuseppe De Giacomo and Yves Lespérance}, doi = {10.1016/j.artint.2025.104407}, journal = {Artificial Intelligence}, month = {11}, pages = {104407}, shortjournal = {Artif. Intell.}, title = {Abstracting situation calculus action theories}, volume = {348}, year = {2025}, }
• (2025). Towards optimal subsidy bounds for envy-freeable allocations. AIJ, 348, 104406. (www)

  @article{ , abstract = {We study the fair division of indivisible items with subsidies among n agents, where the absolute marginal valuation of each item is at most one. Under monotone nondecreasing valuations (where each item is a good), Brustle et al. [9] demonstrated that a maximum subsidy of 2 ( n − 1 ) and a total subsidy of 2 ( n − 1 ) 2 are sufficient to guarantee the existence of an envy-freeable allocation. In this paper, we improve upon these bounds, even in a wider model. Namely, we show that, given an EF1 allocation, we can compute in polynomial time an envy-free allocation with a subsidy of at most n − 1 per agent and a total subsidy of at most n ( n − 1 ) / 2 . Moreover, when the valuations are monotone nondecreasing, we provide a polynomial-time algorithm that computes an envy-free allocation with a subsidy of at most n − 1.5 per agent and a total subsidy of at most ( n 2 − n − 1 ) / 2 .}, archive = {J_AIJ}, author = {Yasushi Kawase and Kazuhisa Makino and Hanna Sumita and Akihisa Tamura and Makoto Yokoo}, doi = {10.1016/j.artint.2025.104406}, journal = {Artificial Intelligence}, month = {11}, pages = {104406}, shortjournal = {Artif. Intell.}, title = {Towards optimal subsidy bounds for envy-freeable allocations}, volume = {348}, year = {2025}, }
• (2025). Local-MIP: Efficient local search for mixed integer programming. AIJ, 348, 104405. (www)

  @article{ , abstract = {Mixed Integer Programming (MIP) is a fundamental model in operations research with broad industrial applications. Local search is a powerful methodology for solving complex optimization problems; however, the development of local search algorithms for MIP still needs exploration. In this work, we propose Local-MIP , an efficient local search algorithm tailored for MIP that integrates novel operators and employs a two-mode architecture to adaptively apply operators based on the current solution's feasibility. For the feasible mode, we propose the lift move operator and a corresponding lift process to improve the objective value while maintaining feasibility. For the infeasible mode, we propose the breakthrough move and mixed tight move operators to respectively optimize the objective function and satisfy constraints. To apply operators intelligently, we develop a dynamic weighting scheme that balances the priorities of the objective function and constraints. Furthermore, we propose a two-level scoring function structure that hierarchically selects operations, guiding the search toward high-quality feasible solutions. Experiments are conducted on public benchmarks to compare Local-MIP with state-of-the-art MIP solvers in finding high-quality solutions. The results show that Local-MIP significantly outperforms CPLEX , HiGHS , SCIP , and Feasibility Jump while remaining competitive with the commercial solver Gurobi on challenging problems within short time limits. Moreover, Local-MIP establishes 10 new records on MIPLIB open instances.}, archive = {J_AIJ}, author = {Peng Lin and Shaowei Cai and Mengchuan Zou and Jinkun Lin}, doi = {10.1016/j.artint.2025.104405}, journal = {Artificial Intelligence}, month = {11}, pages = {104405}, shortjournal = {Artif. Intell.}, title = {Local-MIP: Efficient local search for mixed integer programming}, volume = {348}, year = {2025}, }
• (2025). Centralized training with hybrid execution in multi-agent reinforcement learning via predictive observation imputation. AIJ, 348, 104404. (www)

  @article{ , abstract = {We study hybrid execution in multi-agent reinforcement learning (MARL), a paradigm where agents aim to complete cooperative tasks with arbitrary communication levels at execution time by taking advantage of information-sharing among the agents. Under hybrid execution, the communication level can range from a setting in which no communication is allowed between agents (fully decentralized), to a setting featuring full communication (fully centralized), but the agents do not know beforehand which communication level they will encounter at execution time. We contribute MARO, an approach that makes use of an auto-regressive predictive model, trained in a centralized manner, to estimate missing agents' observations at execution time. We evaluate MARO on standard scenarios and extensions of previous benchmarks tailored to emphasize the impact of partial observability in MARL. Experimental results show that our method consistently outperforms relevant baselines, allowing agents to act with faulty communication while successfully exploiting shared information.}, archive = {J_AIJ}, author = {Pedro P. Santos and Diogo S. Carvalho and Miguel Vasco and Alberto Sardinha and Pedro A. Santos and Ana Paiva and Francisco S. Melo}, doi = {10.1016/j.artint.2025.104404}, journal = {Artificial Intelligence}, month = {11}, pages = {104404}, shortjournal = {Artif. Intell.}, title = {Centralized training with hybrid execution in multi-agent reinforcement learning via predictive observation imputation}, volume = {348}, year = {2025}, }
• (2025). Algebras of actions in an agent's representations of the world. AIJ, 348, 104403. (www)

  @article{ , abstract = {Learning efficient representations allows robust processing of data, data that can then be generalised across different tasks and domains, and it is thus paramount in various areas of Artificial Intelligence, including computer vision, natural language processing and reinforcement learning, among others. Within the context of reinforcement learning, we propose in this paper a mathematical framework to learn representations by extracting the algebra of the transformations of worlds from the perspective of an agent. As a starting point, we use our framework to reproduce representations from the symmetry-based disentangled representation learning (SBDRL) formalism proposed by [1] and prove that, although useful, they are restricted to transformations that respond to the properties of algebraic groups. We then generalise two important results of SBDRL –the equivariance condition and the disentangling definition– from only working with group-based symmetry representations to working with representations capturing the transformation properties of worlds for any algebra, using examples common in reinforcement learning and generated by an algorithm that computes their corresponding Cayley tables. Finally, we combine our generalised equivariance condition and our generalised disentangling definition to show that disentangled sub-algebras can each have their own individual equivariance conditions, which can be treated independently, using category theory. In so doing, our framework offers a rich formal tool to represent different types of symmetry transformations in reinforcement learning, extending the scope of previous proposals and providing Artificial Intelligence developers with a sound foundation to implement efficient applications.}, archive = {J_AIJ}, author = {Alexander Dean and Eduardo Alonso and Esther Mondragón}, doi = {10.1016/j.artint.2025.104403}, journal = {Artificial Intelligence}, month = {11}, pages = {104403}, shortjournal = {Artif. Intell.}, title = {Algebras of actions in an agent's representations of the world}, volume = {348}, year = {2025}, }
• (2025). Enhancing cooperativity in controlled query evaluation over ontologies. AIJ, 348, 104402. (www)

  @article{ , abstract = {Controlled Query Evaluation (CQE) is a methodology designed to maintain confidentiality by either rejecting specific queries or adjusting responses to safeguard sensitive information. In this investigation, our focus centers on CQE within Description Logic ontologies, aiming to ensure that queries are answered truthfully as long as possible before resorting to deceptive responses, a cooperativity property which is called the “longest honeymoon”. Our work introduces new semantics for CQE, denoted as MC-CQE, which enjoys the longest honeymoon property and outperforms previous methodologies in terms of cooperativity. We study the complexity of query answering in this new framework for ontologies expressed in the Description Logic DL-Lite R . Specifically, we establish data complexity results under different maximally cooperative semantics and for different classes of queries. Our results identify both tractable and intractable cases. In particular, we show that the evaluation of Boolean unions of conjunctive queries is the same under all the above semantics and its data complexity is in . This result makes query answering amenable to SQL query rewriting. However, this favorable property does not extend to open queries, even with a restricted query language limited to conjunctions of atoms. While, in general, answering open queries in the MC-CQE framework is intractable, we identify a sub-family of semantics under which answering full conjunctive queries is tractable.}, archive = {J_AIJ}, author = {Piero Bonatti and Gianluca Cima and Domenico Lembo and Francesco Magliocca and Lorenzo Marconi and Riccardo Rosati and Luigi Sauro and Domenico Fabio Savo}, doi = {10.1016/j.artint.2025.104402}, journal = {Artificial Intelligence}, month = {11}, pages = {104402}, shortjournal = {Artif. Intell.}, title = {Enhancing cooperativity in controlled query evaluation over ontologies}, volume = {348}, year = {2025}, }
• (2025). BATED: Learning fair representation for pre-trained language models via biased teacher-guided disentanglement. AIJ, 348, 104401. (www)

  @article{ , abstract = {With the rapid development of Pre-trained Language Models (PLMs) and their widespread deployment in various real-world applications, social biases of PLMs have attracted increasing attention, especially the fairness of downstream tasks, which potentially affects the development and stability of society. Among existing debiasing methods, intrinsic debiasing methods are not necessarily effective when applied to downstream tasks, and the downstream fine-tuning process may introduce new biases or catastrophic forgetting. Most extrinsic debiasing methods rely on sensitive attribute words as prior knowledge to supervise debiasing training. However, it is difficult to collect sensitive attribute information of real data due to privacy and regulation. Moreover, limited sensitive attribute words may lead to inadequate debiasing training. To this end, this paper proposes a debiasing method to learn fair representation for PLMs via B i A sed TE acher-guided D isentanglement (called BATED ). Specific to downstream tasks, BATED performs debiasing training under the guidance of a biased teacher model rather than relying on sensitive attribute information of the training data. First, we leverage causal contrastive learning to train a task-agnostic general biased teacher model. We then employ Variational Auto-Encoder (VAE) to disentangle the PLM-encoded representation into the fair representation and the biased representation. The Biased representation is further decoupled via biased teacher-guided disentanglement, while the fair representation learn downstream tasks. Therefore, BATED guarantees the performance of downstream tasks while improving the fairness. Experimental results on seven PLMs testing three downstream tasks demonstrate that BATED outperforms the state-of-the-art overall in terms of fairness and performance on downstream tasks.}, archive = {J_AIJ}, author = {Yingji Li and Mengnan Du and Rui Song and Mu Liu and Ying Wang}, doi = {10.1016/j.artint.2025.104401}, journal = {Artificial Intelligence}, month = {11}, pages = {104401}, shortjournal = {Artif. Intell.}, title = {BATED: Learning fair representation for pre-trained language models via biased teacher-guided disentanglement}, volume = {348}, year = {2025}, }
• (2025). On preference learning based on sequential bayesian optimization with pairwise comparison. AIJ, 348, 104400. (www)

  @article{ , abstract = {User preference learning is generally a hard problem. Individual preferences are typically unknown even to users themselves, while the space of choices is infinite. Here we study user preference learning from information-theoretic perspective. We model preference learning as a system with two interacting sub-systems, one representing a user with his/her preferences and another one representing an agent that has to learn these preferences. The user with his/her behavior is modeled by a parametric preference function. To efficiently learn the preferences and reduce search space quickly, we propose the agent that interacts with the user to collect the most informative data for learning. The agent presents two proposals to the user for evaluation, and the user rates them based on his/her preference function. We show that the optimum agent strategy for data collection and preference learning is a result of maximin optimization of the normalized weighted Kullback-Leibler (KL) divergence between true and agent-assigned predictive user response distributions. The resulting value of the KL-divergence, which we also call of a remaining system uncertainty (RSU), provides an efficient performance metric in the absence of the ground truth. This metric characterizes how well the agent can predict user and, thus, the quality of the underlying learned user (preference) model. Our proposed agent comprises sequential mechanisms for user model inference and proposal generation. To infer the user model (preference function), Bayesian approximate inference is used in the agent. The data collection strategy is to generate proposals, responses to which help resolving uncertainty associated with prediction of the user responses the most. The efficiency of our approach is validated by numerical simulations. Also a real-life example of preference learning application is provided.}, archive = {J_AIJ}, author = {Tanya Ignatenko and Kirill Kondrashov and Marco Cox and Bert de Vries}, doi = {10.1016/j.artint.2025.104400}, journal = {Artificial Intelligence}, month = {11}, pages = {104400}, shortjournal = {Artif. Intell.}, title = {On preference learning based on sequential bayesian optimization with pairwise comparison}, volume = {348}, year = {2025}, }
• (2025). Choosing abstraction levels for model-based software debugging: A theoretical and empirical analysis for spreadsheet programs. AIJ, 348, 104399. (www)

  @article{ , abstract = {Model-based diagnosis is a generally applicable, principled approach to the systematic debugging of a wide range of system types such as circuits, knowledge bases, physical devices, or software. Based on a formal description of the system, it enables precise and deterministic reasoning about potential faults responsible for observed misbehavior. In software, such a formal system description can often even be extracted from the buggy program fully automatically. As logical reasoning is central to diagnosis, the performance of model-based debuggers is largely influenced by reasoning efficiency, which in turn depends on the complexity and expressivity of the system description. Since highly detailed models capturing exact semantics often exceed the capabilities of current reasoning tools, researchers have proposed more abstract representations. In this work, we thoroughly analyze system modeling techniques with a focus on fault localization in spreadsheets—one of the most widely used end-user programming paradigms. Specifically, we present three constraint model types characterizing spreadsheets at different abstraction levels, show how to extract them automatically from faulty spreadsheets, and provide theoretical and empirical investigations of the impact of abstraction on both diagnostic output and computational performance. Our main conclusions are that (i) for the model types, there is a trade-off between the conciseness of generated fault candidates and computation time, (ii) the exact model is often impractical, and (iii) a new model based on qualitative reasoning yields the same solutions as the exact one in up to more than half the cases while being orders of magnitude faster. Due to their ability to restrict the solution space in a sound way, the explored model-based techniques, rather than being used as standalone approaches, are expected to realize their full potential in combination with iterative sequential diagnosis or indeterministic but more performant statistical debugging methods.}, archive = {J_AIJ}, author = {Patrick Rodler and Birgit Hofer and Dietmar Jannach and Iulia Nica and Franz Wotawa}, doi = {10.1016/j.artint.2025.104399}, journal = {Artificial Intelligence}, month = {11}, pages = {104399}, shortjournal = {Artif. Intell.}, title = {Choosing abstraction levels for model-based software debugging: A theoretical and empirical analysis for spreadsheet programs}, volume = {348}, year = {2025}, }
• (2025). Argus: Programming with communication protocols in a belief-desire-intention architecture. AIJ, 348, 104398. (www)

  @article{ , abstract = {Protocols model multiagent systems (MAS) by capturing the communications between its agents. Belief-Desire-Intention (BDI) architectures provide an attractive way for organizing an agent in terms of cognitive concepts. Current BDI approaches, however, lack adequate support for engineering protocol-based agents. We describe Argus, an approach that melds recent advances in flexible, declarative communication protocols with BDI architectures. For concreteness, we adopt Jason as an exemplar of the BDI paradigm and show how to support protocol-based reasoning in it. Specifically, Argus contributes (1) a novel architecture and formal operational semantics combining protocols and BDI; (2) a code generation-based programming model that guides the implementation of agents; and (3) integrity checking for incoming and outgoing messages that help ensure that the agents are well-behaved. The Argus conceptual architecture builds quite naturally on top of Jason. Thus, Argus enables building more flexible multiagent systems while using a BDI architecture than is currently possible.}, archive = {J_AIJ}, author = {Samuel H. Christie V and Munindar P. Singh and Amit K. Chopra}, doi = {10.1016/j.artint.2025.104398}, journal = {Artificial Intelligence}, month = {11}, pages = {104398}, shortjournal = {Artif. Intell.}, title = {Argus: Programming with communication protocols in a belief-desire-intention architecture}, volume = {348}, year = {2025}, }
• (2025). Social behavior as a key to learning-based multi-agent pathfinding dilemmas. AIJ, 348, 104397. (www)

  @article{ , abstract = {The Multi-agent Path Finding (MAPF) problem involves finding collision-free paths for a team of agents in a known, static environment, with important applications in warehouse automation, logistics, or last-mile delivery. To meet the needs of these large-scale applications, current learning-based methods often deploy the same fully trained, decentralized network to all agents to improve scalability. However, such parameter sharing typically results in homogeneous behaviors among agents, which may prevent agents from breaking ties around symmetric conflict (e.g., bottlenecks) and might lead to live-/deadlocks. In this paper, we propose SYLPH, a novel learning-based MAPF framework aimed to mitigate the adverse effects of homogeneity by allowing agents to learn and dynamically select different social behaviors (akin to individual, dynamic roles), without affecting the scalability offered by parameter sharing. Specifically, SYLPH offers a novel hierarchical mechanism by introducing Social Value Orientation (SVO) as a temporally extended latent variable that plays a central role in both policy generation and reward assignment. To support this hierarchical decision-making process, we introduce Social-aware Multi-Policy PPO (SMP3O), a reinforcement learning method that ensures stable and effective training through a mechanism for the cross-utilization of advantages. Moreover, we design an SVO-based learning tie-breaking algorithm, allowing agents to proactively avoid collisions, rather than relying solely on post-processing techniques. As a result of this hierarchical decision-making and exchange of social preferences, SYLPH endows agents with the ability to reason about the MAPF task through more latent spaces and nuanced contexts, leading to varied responses that can help break ties around symmetric conflicts. Our comparative experiments show that SYLPH achieves state-of-the-art performance, surpassing other learning-based MAPF planners in random, room-like, and maze-like maps, while our ablation studies demonstrate the advantages of each component in SYLPH. We finally experimentally validate our trained policies on hardware in three types of maps, showing how SYLPH allows agents to find high-quality paths under real-life conditions. Our code and videos are available at: marmotlab.github.io/mapf_sylph .}, archive = {J_AIJ}, author = {Chengyang He and Tanishq Duhan and Parth Tulsyan and Patrick Kim and Guillaume Sartoretti}, doi = {10.1016/j.artint.2025.104397}, journal = {Artificial Intelligence}, month = {11}, pages = {104397}, shortjournal = {Artif. Intell.}, title = {Social behavior as a key to learning-based multi-agent pathfinding dilemmas}, volume = {348}, year = {2025}, }
• (2025). MATE: Masked optimal transport with dynamic selection for partial label graph learning. AIJ, 348, 104396. (www)

  @article{ , abstract = {This paper investigates the problem of partial label graph learning, in which every graph is associated with a set of candidate labels. Previous methods for weakly supervised graph classification often provide pseudo-labels for graph samples that could be overconfident and biased towards the dominant classes, thus resulting in substantial error accumulation. In this paper, we introduce a new framework named M asked Optim a l T ransport with Dynamic S e lection (MATE) for partial label graph learning, which improves the quality of graph assignments from the perspectives of class balancing and uncertainty mining. In particular, our MATE masks probabilities out of candidate sets and then adopts optimal transport to optimize the assignments without class biases. This design is based on the assumption that the true label distribution is class-balanced or nearly balanced, which is common in various training datasets and real-world scenarios. To further reduce potential noise, we propose a novel scoring metric termed partial energy discrepancy (PED) to evaluate the uncertainty of assignments, and then introduce a dynamic selection strategy that modifies the sample-specific thresholds via momentum updating. Finally, these samples are divided into three levels, i.e., confident, less-confident, and unconfident and each group is trained separately in our collaborative optimization framework. Extensive experiments on various benchmarks demonstrate the superiority of our MATE compared to various state-of-the-art baselines.}, archive = {J_AIJ}, author = {Yiyang Gu and Binqi Chen and Zihao Chen and Ziyue Qiao and Xiao Luo and Junyu Luo and Zhiping Xiao and Wei Ju and Ming Zhang}, doi = {10.1016/j.artint.2025.104396}, journal = {Artificial Intelligence}, month = {11}, pages = {104396}, shortjournal = {Artif. Intell.}, title = {MATE: Masked optimal transport with dynamic selection for partial label graph learning}, volume = {348}, year = {2025}, }
• (2025). Interpreting capsule networks for image classification by routing path visualization. AIJ, 348, 104395. (www)

  @article{ , abstract = {Artificial neural networks are popular for computer vision as they often give state-of-the-art performance, but are difficult to interpret because of their complexity. This black box modeling is especially troubling when the application concerns human well-being such as in medical image analysis or autonomous driving. In this work, we propose a technique called routing path visualization for capsule networks, which reveals how much of each region in an image is routed to each capsule. In turn, this technique can be used to interpret the entity that a given capsule detects, and speculate how the network makes a prediction. We demonstrate our new visualization technique on several real world datasets. Experimental results suggest that routing path visualization can precisely localize the predicted class from an image, even though the capsule networks are trained using just images and their respective class labels, without additional information defining the location of the class in the image.}, archive = {J_AIJ}, author = {Amanjot Bhullar and Michael Czomko and R. Ayesha Ali and Douglas L. Welch}, doi = {10.1016/j.artint.2025.104395}, journal = {Artificial Intelligence}, month = {11}, pages = {104395}, shortjournal = {Artif. Intell.}, title = {Interpreting capsule networks for image classification by routing path visualization}, volume = {348}, year = {2025}, }
• (2025). Relaxed core stability in hedonic games. AIJ, 348, 104394. (www)

  @article{ , abstract = {The core is a well-known and fundamental notion of stability in games intended to model coalition formation such as hedonic games: an outcome is core stable if there exists no blocking coalition , i.e., no set of agents that may profit by forming a coalition together. The fact that the cardinality of a blocking coalition, i.e., the number of deviating agents that have to coordinate themselves, can be arbitrarily high, and the fact that agents may benefit only by a tiny amount from their deviation, while they could incur in a higher cost for deviating, suggest that the core is not able to suitably model practical scenarios in large and highly distributed multi-agent systems. For this reason, we consider relaxed core stable outcomes where the notion of permissible deviations is modified along two orthogonal directions: the former takes into account the size q of the deviating coalition, and the latter the amount of utility gain, in terms of a multiplicative factor k , for each member of the deviating coalition. These changes result in two different notions of stability, namely, the q-size core and k-improvement core . We consider fractional hedonic games, that is a well-known subclass of hedonic games for which core stable outcomes are not guaranteed to exist and it is computationally hard to decide non-emptiness of the core; we investigate these relaxed concepts of stability with respect to their existence, computability and performance in terms of price of anarchy and price of stability, by providing in many cases tight or almost tight bounds. Interestingly, the considered relaxed notions of core also possess the appealing property of recovering, in some notable cases, the convergence, the existence and the possibility of computing stable solutions in polynomial time.}, archive = {J_AIJ}, author = {Angelo Fanelli and Gianpiero Monaco and Luca Moscardelli}, doi = {10.1016/j.artint.2025.104394}, journal = {Artificial Intelligence}, month = {11}, pages = {104394}, shortjournal = {Artif. Intell.}, title = {Relaxed core stability in hedonic games}, volume = {348}, year = {2025}, }
• (2025). Provably efficient information-directed sampling algorithms for multi-agent reinforcement learning. AIJ, 348, 104392. (www)

  @article{ , abstract = {This work designs and analyzes a novel set of algorithms for multi-agent reinforcement learning (MARL) based on the principle of information-directed sampling (IDS). These algorithms draw inspiration from foundational concepts in information theory, and are proven to be sample efficient in MARL settings such as two-player zero-sum Markov games (MGs) and multi-player general-sum MGs. For episodic two-player zero-sum MGs, we present three sample-efficient algorithms for learning Nash equilibrium. The basic algorithm, referred to as MAIDS , employs an asymmetric learning structure where the max-player first solves a minimax optimization problem based on the joint information ratio of the joint policy, and the min-player then minimizes the marginal information ratio with the max-player's policy fixed. Theoretical analyses show that it achieves a Bayesian regret of O ˜ ( K ) for K episodes. To reduce the computational load of MAIDS , we develop an improved algorithm called Reg-MAIDS , which has the same Bayesian regret bound while enjoying less computational complexity. Moreover, by leveraging the flexibility of IDS principle in choosing the learning target, we propose two methods for constructing compressed environments based on rate-distortion theory, upon which we develop an algorithm Compressed-MAIDS wherein the learning target is a compressed environment. Finally, we extend Reg-MAIDS to multi-player general-sum MGs and prove that it can learn either the Nash equilibrium or coarse correlated equilibrium in a sample-efficient manner.}, archive = {J_AIJ}, author = {Qiaosheng Zhang and Chenjia Bai and Shuyue Hu and Zhen Wang and Xuelong Li}, doi = {10.1016/j.artint.2025.104392}, journal = {Artificial Intelligence}, month = {11}, pages = {104392}, shortjournal = {Artif. Intell.}, title = {Provably efficient information-directed sampling algorithms for multi-agent reinforcement learning}, volume = {348}, year = {2025}, }
• (2025). On the design of truthful mechanisms for the capacitated facility location problem with two and more facilities. AIJ, 348, 104390. (www)

  @article{ , abstract = {In this paper, we explore the Mechanism Design aspects of the m -Capacitated Facility Location Problem ( m -CFLP) on a line, focusing on two frameworks. In the first framework, the number of facilities is arbitrary, all facilities share the same capacity, and the number of agents matches the total capacity of the facilities. In the second framework, we need to locate two facilities, each with a capacity equal to at least half the number of agents. For both frameworks, we propose truthful mechanisms with bounded approximation ratios in terms of Social Cost (SC) and Maximum Cost (MC). When m > 2 , our results stand in contrast to the impossibility results known for the classical m -Facility Location Problem, where capacity constraints are absent. Moreover, all the proposed mechanisms are optimal with respect to MC and either optimal or near-optimal with respect to the SC among anonymous mechanisms. We then establish lower bounds on the approximation ratios that any truthful and deterministic mechanism achieves with respect to SC and MC for both frameworks. Lastly, we run several numerical experiments to empirically evaluate the performances of our mechanisms with respect to the SC or the MC. Our empirical analysis shows that our proposed mechanisms outperform all previously proposed mechanisms applicable in this setting.}, archive = {J_AIJ}, author = {Gennaro Auricchio and Zihe Wang and Jie Zhang}, doi = {10.1016/j.artint.2025.104390}, journal = {Artificial Intelligence}, month = {11}, pages = {104390}, shortjournal = {Artif. Intell.}, title = {On the design of truthful mechanisms for the capacitated facility location problem with two and more facilities}, volume = {348}, year = {2025}, }

AMC - 57

• (2026). The list r-hued coloring of p5-free graph. AMC, 511, 129742. (www)

  @article{ , abstract = {Given a list L of graph G , an ( L , r ) -coloring of graph G is a proper coloring such that the color of vertex v belongs to its list L ( v ) , and each vertex of degree d G ( v ) is adjcent to vertices with at least min { r , d G ( v ) } different colors. The list r -hued chromatic number, denoted by χ L , r ( G ) , is the smallest integer k such that for any k -list L of G , G has an ( L , r ) -coloring. We prove the following: ( i ) If G = G ( U , V ) is a connected P 5 -free bipartite graph and min { | U | , | V | } = s , then χ L , r ( G ) ≤ max { 2 r , s + 1 } . ( i i ) If G is a connected P 5 -free graph and r ≤ 3 , then χ L , r ( G ) ≤ r χ L ( G ) .}, archive = {J_AMC}, author = {Xuanhe Jia and Fengxia Liu and Baiyun Ji and Zihang Zhao}, doi = {10.1016/j.amc.2025.129742}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129742}, shortjournal = {Appl. Math. Comput.}, title = {The list r-hued coloring of p5-free graph}, volume = {511}, year = {2026}, }
• (2026). Stability analysis of stochastic McKean–Vlasov equations with discrete observation. AMC, 511, 129740. (www)

  @article{ , abstract = {This paper explores the stability issue of stochastic McKean–Vlasov equations (SMVEs) through an innovative method: stabilization via stochastic delay feedback control with discrete observation. Unlike conventional techniques, this approach leverages historical states instead of solely relying on the current state, setting it apart from traditional stochastic feedback controls. The study examines how the diffusion term contributes to enhancing system stability despite the presence of random fluctuations and delays, guaranteeing both p -th moment exponential stability and almost sure exponential stability under a specific delay threshold δ * . The primary contributions of this work include introducing a novel stability analysis framework utilizing stochastic delay feedback control, constructing Lyapunov functions that incorporate both state and distribution, and providing insights into asymptotic and moment exponential stability. Although identifying the optimal delay δ * remains a practical challenge, the theoretical foundation laid in this study offers valuable guidance for real-world applications.}, archive = {J_AMC}, author = {Yicheng Liu and Quanxin Zhu}, doi = {10.1016/j.amc.2025.129740}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129740}, shortjournal = {Appl. Math. Comput.}, title = {Stability analysis of stochastic McKean–Vlasov equations with discrete observation}, volume = {511}, year = {2026}, }
• (2026). Collaborated risk perception against multiple epidemics in a multiplayer network. AMC, 511, 129738. (www)

  @article{ , abstract = {The concurrent outbreak of epidemics has become a great threat to public health. An important problem is how individuals can protect themselves as the mandated protective measures are relaxed. Previous studies developed various models to investigate the correlated spreading dynamics of concurrent epidemics and the protective measures against them. However, a critical oversight remains that people’s risk perceptions on multiple epidemics are also correlated or even collaborated. In this paper, we build an SS-IS-SI-II coupled model in a multilayer network to describe two concurrent epidemics, integrating collaborated risk perception and spontaneous social distancing as individuals’ self-protection. Moreover, an adjustable coefficient is proposed to describe different levels of inter-epidemic correlations (competition/independence/cooperation). It is found that increasing the levels of inter-epidemic correlation will increase the infected density of both epidemics. Collaborated risk perception is generally more effective in reducing infections across different levels of inter-epidemic correlation, compared with independent risk perception. But its effect is dependent. For one epidemic, when its infectivity is very high or the infectivity of the other epidemic is very low, the effect of collaborated risk perception will be largely reduced. Based on the model, we further investigate the minimum social distancing required to contain epidemics under different conditions, and the priority on different layers is also explored. This research extends existing literature on co-evolutional epidemic dynamics, and lay a foundation to model correlated risk perceptions against epidemics. The results provide implications for individuals to take self-protection against concurrent epidemics.}, archive = {J_AMC}, author = {Yahong Chen and He Huang}, doi = {10.1016/j.amc.2025.129738}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129738}, shortjournal = {Appl. Math. Comput.}, title = {Collaborated risk perception against multiple epidemics in a multiplayer network}, volume = {511}, year = {2026}, }
• (2026). Superconvergent methods for solving two-dimensional hammerstein integral equations. AMC, 511, 129737. (www)

  @article{ , abstract = {In this paper, we introduce the superconvergent degenerate kernel method and the superconvergent Nyström method for the numerical solution of two-dimensional Hammerstein integral equations of the second kind. By employing piecewise polynomial interpolation of degree r , we prove that, under symmetry conditions on both the triangulation and the interpolation nodes, convergence orders of 2 r + 3 and 2 r + 4 are achieved for the approximate solutions and their iterated versions, respectively. Furthermore, we discuss computational aspects related to the construction of the corresponding nonlinear systems, and we present numerical examples to illustrate the theoretical results obtained.}, archive = {J_AMC}, author = {M. Sennour and D. Sbibih and M. Tahrichi}, doi = {10.1016/j.amc.2025.129737}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129737}, shortjournal = {Appl. Math. Comput.}, title = {Superconvergent methods for solving two-dimensional hammerstein integral equations}, volume = {511}, year = {2026}, }
• (2026). Integrating emotion and expectation improves cooperation. AMC, 511, 129736. (www)

  @article{ , abstract = {Cooperation is a key issue attracting widespread attention across various fields. Emotions and expectations jointly participate in the decision-making process, and are key factors influencing the evolution of cooperation. Therefore, this paper proposes a game evolution model that considers the dual influences of emotions and expectations. In the model, a player’s initial strategy depends on their emotions. Subsequently, expectations influence the decision-making process, altering the player’s initial strategy to the current strategy. Specifically, high expectations lead players to maintain their initial strategy, whereas low expectations prompt them to change strategies. Furthermore, the paper analyzes the evolution of cooperation in the prisoner’s dilemma and the influence of expectations. Simulation results show that under high betrayal temptation, players with loneliness and defection would change their initial strategy to cooperation due to low expectations, while those players who initially gather with cooperation strategies maintain the cooperation strategies due to higher expectations, ultimately increasing the cooperation fraction and payoff of the population. Therefore, expectations effectively enhance the cooperation level and payoff of the population. When the scale of individuals changing their strategies based on expectations (ICSE) is large, the effect of expectations on enhancing cooperation and payoff is more pronounced. Frequent consideration of anticipated initial strategy changes also yields the same effect. Additionally, mechanisms of the influence of emotion on payoff, direct emotion interaction and the influence of strategy on emotion collectively contribute to enhancing the significantly evolutionary advantage of friendly emotions and cooperation strategies. This paper contributes to a deeper understanding of the role of expectations in the evolution of emotions and cooperation, laying the groundwork for enhancing social cooperation through expectation management.}, archive = {J_AMC}, author = {Wen Lu and Shu Liang}, doi = {10.1016/j.amc.2025.129736}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129736}, shortjournal = {Appl. Math. Comput.}, title = {Integrating emotion and expectation improves cooperation}, volume = {511}, year = {2026}, }
• (2026). On sketch-and-project methods for solving tensor equations. AMC, 511, 129735. (www)

  @article{ , abstract = {We propose a regular sketch-and-project method for solving linear tensor equations based on the t-product and present its equivalent Fourier domain version, along with several special cases corresponding to existing classical matrix equation methods. Furthermore, we extend this framework via a hierarchical approach to solve generalized Sylvester tensor equations. All the methods are proved to converge linearly in expectation. Finally, numerical experiments demonstrate the efficiency and effectiveness of the proposed approach.}, archive = {J_AMC}, author = {Ling Tang and Yanjun Zhang and Hanyu Li}, doi = {10.1016/j.amc.2025.129735}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129735}, shortjournal = {Appl. Math. Comput.}, title = {On sketch-and-project methods for solving tensor equations}, volume = {511}, year = {2026}, }
• (2026). Sampling patterns for zernike-like bases in non-standard geometries. AMC, 511, 129727. (www)

  @article{ , abstract = {Zernike polynomials are widely used in optics and ophthalmology due to their direct connection to classical optical aberrations. While orthogonal on the unit disk, their application to discrete data or non-circular domains–such as ellipses, annuli, and hexagons–presents challenges in terms of numerical stability and accuracy. In this work, we extend Zernike-like orthogonal functions to these non-standard geometries using diffeomorphic mappings and construct sampling patterns that preserve favorable numerical conditioning. We provide theoretical bounds for the condition numbers of the resulting collocation matrices and validate them through extensive numerical experiments. As a practical application, we demonstrate accurate wavefront interpolation and reconstruction in segmented mirror telescopes composed of hexagonal facets. Our results show that appropriately transferred sampling configurations, especially Optimal Concentric Sampling and Lebesgue points , allow stable high-order interpolation and effective wavefront modeling in complex optical systems. Moreover, the Optimal Concentric Samplings can be computed with an explicit expression, which is a significant advantage in practice.}, archive = {J_AMC}, author = {S. Díaz-Elbal and A. Martínez-Finkelshtein and D. Ramos-López}, doi = {10.1016/j.amc.2025.129727}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129727}, shortjournal = {Appl. Math. Comput.}, title = {Sampling patterns for zernike-like bases in non-standard geometries}, volume = {511}, year = {2026}, }
• (2026). Optimal information spreading strategy for containing epidemic spreading on higher-order multiplex networks. AMC, 511, 129725. (www)

  @article{ , abstract = {When an epidemic spreads through a population, related information also spreads concurrently, prompting individuals to adopt protective behaviours (e.g., washing hands). Collective behaviour has been shown to play a critical role in shaping the dynamics of epidemic spreading, and higher-order networks offer a natural framework to describe such group interactions in social contact networks. Yet, the interplay between epidemic and information dynamics on higher-order structures is not fully understood, further limiting our understanding of the optimal information spreading strategy for containing epidemic spreading.In this study, we first construct a higher-order multiplex network framework based on simplicial complexes. Then, a coevolutionary spreading model is proposed, integrating epidemic spreading and information spreading on simplicial complexes. The epidemic spreads through both lower-order (pairwise) and higher-order (group) interactions, while information spreads through lower-order interactions in a degree-preferential manner. Using an extended Microscopic Markov Chain Approach, we analytically derive the dynamical equations of the system and compute the basic reproduction number using the next-generation matrix method. Finally, we conduct extensive numerical simulations of the spreading process across various parameter regimes. Our results demonstrate the role of higher-order infections in promoting epidemics. Although information spreading generally suppresses the spread of most epidemics, it can paradoxically enhance the spread of certain epidemics with a very low spreading capacity. Increases in the recovery probabilities of both the disease and the information can weaken the promoting effect of higher-order infection and enhance the suppressive effect of the information. For certain epidemics with weak spreading capabilities but strong recovery capabilities, the spread of information can completely suppress the outbreak of the disease, while the enhancement of higher-order infections can promote the outbreak of these diseases. By analysing the effects of different information spreading strategies on epidemic spreading, we find that the optimal strategy for containing the epidemic is to allow information to spread without degree preference.}, archive = {J_AMC}, author = {Jiayi Song and Wenjie Li and Yunzhu Xiao and Ling Chen and Chun Yang and Li Qi and Wei Wang}, doi = {10.1016/j.amc.2025.129725}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129725}, shortjournal = {Appl. Math. Comput.}, title = {Optimal information spreading strategy for containing epidemic spreading on higher-order multiplex networks}, volume = {511}, year = {2026}, }
• (2026). On the fairness and cooperation among free-riders. AMC, 511, 129724. (www)

  @article{ , abstract = {We consider the free-rider problem in peer-to-peer (P2P) systems, where agents enjoy the group effort without contributing their share. We introduce the Free-Rider Game (FRG), a non-cooperative game incorporating a fairness-aware profit allocation rule based on the Robin Hood index. We show that FRG admits strong structural properties. First, making a non-zero contribution is a dominant strategy for any player. Second, a player contributes positively whenever at least one other player does so. Third, FRG admits a unique Nash equilibrium in which each player contributes the fullest, eliminating free riding. Fourth, equilibrium outcomes are proportionally fair, ensuring balanced allocation across agents. Finally, FRG guarantees full participation by embedding fairness directly into the payoff structure, differentiating it from classical public goods games, which often yield zero or partial contributions.}, archive = {J_AMC}, author = {Avadh Kishor}, doi = {10.1016/j.amc.2025.129724}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129724}, shortjournal = {Appl. Math. Comput.}, title = {On the fairness and cooperation among free-riders}, volume = {511}, year = {2026}, }
• (2026). Equilibrium analysis of edge-heterogeneous binary network games. AMC, 511, 129723. (www)

  @article{ , abstract = {Recent studies on the equilibrium of binary network games have primarily focused on scenarios characterized by agent heterogeneity, where agents exhibit unique attributes but their interactions with different neighbors remain uniform. In this paper, we investigate the edge-heterogeneous binary network game, a more general framework that incorporates heterogeneity into agent interactions. We establish two sufficient equilibrium conditions under asynchronous best-response dynamics from different perspectives. The first condition requires underlying symmetry in interactions between neighboring agents, integrating and generalizing three classical convergence situations in binary network games. The second condition focuses on network balance, positing that equilibrium is achievable if the coordination value network of a game is structurally balanced. Additionally, for games meeting this condition, we develop a method to predict the final state based on initial state information. These results reveal factors that steer edge-heterogeneous binary network games towards equilibrium, providing valuable insights for controlling such highly nonlinear systems. Lastly, we extend the analysis to higher-order network games and propose an equilibrium condition for edge-heterogeneous 2-order network games.}, archive = {J_AMC}, author = {Jiawen Wang and Yangyang Luan and Xiaoqun Wu}, doi = {10.1016/j.amc.2025.129723}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129723}, shortjournal = {Appl. Math. Comput.}, title = {Equilibrium analysis of edge-heterogeneous binary network games}, volume = {511}, year = {2026}, }
• (2026). Theory and numerics of subspace approximation of eigenvalue problems. AMC, 511, 129722. (www)

  @article{ , abstract = {Large-scale eigenvalue problems arise in various fields of science and engineering and demand computationally efficient solutions. In this study, we investigate the subspace approximation for parametric linear eigenvalue problems, aiming to mitigate the computational burden associated with high-fidelity systems. We provide general error estimates under non-simple eigenvalue conditions, establishing some theoretical foundations for understanding the convergence behavior of subspace approximations. Numerical examples, including problems with one-dimensional to three-dimensional spatial domain and one-dimensional to two-dimensional parameter domain, are presented to demonstrate the efficacy of reduced basis method in handling parametric variations in boundary conditions and coefficient fields to achieve significant computational savings while maintaining high accuracy, making them promising tools for practical applications in large-scale eigenvalue computations.}, archive = {J_AMC}, author = {Siu Wun Cheung and Youngsoo Choi and Seung Whan Chung and Jean-Luc Fattebert and Coleman Kendrick and Daniel Osei-Kuffuor}, doi = {10.1016/j.amc.2025.129722}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129722}, shortjournal = {Appl. Math. Comput.}, title = {Theory and numerics of subspace approximation of eigenvalue problems}, volume = {511}, year = {2026}, }
• (2026). H-function-based state bounding results of discrete-time delayed systems. AMC, 511, 129721. (www)

  @article{ , abstract = {This paper focuses on the h -function-based state bounding estimation problem for discrete-time nonlinear systems (DTNSs) with time-varying delays and bounded disturbances. First, a direct method based on the system solutions is proposed to provide sufficient conditions, which are composed of simple inequalities and depend on the time delays, to ensure that the state trajectories of the considered system always stay within a polyhedron or converge into it. Second, it is demonstrated that the obtained sufficient conditions are precisely the global h -stability (G h -S) criteria of the considered system when disturbances disappear, and when the initial function is restricted within a certain range, the resulting polyhedron can be considered as h -function-based reachable set estimation of the states. Finally, the applicability of the theoretical results obtained is illustrated through two numerical examples.}, archive = {J_AMC}, author = {Huan Zhang and Xiaona Yang and Tianqiu Yu}, doi = {10.1016/j.amc.2025.129721}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129721}, shortjournal = {Appl. Math. Comput.}, title = {H-function-based state bounding results of discrete-time delayed systems}, volume = {511}, year = {2026}, }
• (2026). Composite sliding mode adaptive tracking control for singularly perturbed semi-markov jump non-linear systems with fault and disturbances. AMC, 511, 129717. (www)

  @article{ , abstract = {This work epitomizes the construction of composite sliding mode-oriented adaptive tracking controller for singularly perturbed semi-Markov jump non-linear systems exposed to multiple disturbances and actuator fault. Typically, a model reference adaptive tracking control algorithm composed of baseline control, enhanced error correction term and adaptive laws is embodied with sliding mode control law to attain robust and rapid tracking performance. On top of that, a disturbance observer is embedded with fault estimator configuration to simultaneously deliver precise evaluations of both exogenous disturbances and faults that influence the system state tracking processes and these assessment are assimilated into the established tracking protocol. Collectively, a robust composite sliding mode-oriented adaptive tracking control strategy with disturbance estimation and fault-tolerance strategies is devised to assure adequate tracking objectives as well as the extended dissipativity performance specifications. After then, by deploying relevant Lyapunov function terms, the necessities for confirming the stochastic stability with preset extended dissipativity performance of the adopted system is detailed into linear matrix inequalities. Furthermore, the reliability of the analysed findings is further affirmed by graphical illustrations attained from numerical simulations.}, archive = {J_AMC}, author = {N. Shobana and R. Sakthivel and O.M. Kwon and M. Sankar and M. Asha Safana}, doi = {10.1016/j.amc.2025.129717}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129717}, shortjournal = {Appl. Math. Comput.}, title = {Composite sliding mode adaptive tracking control for singularly perturbed semi-markov jump non-linear systems with fault and disturbances}, volume = {511}, year = {2026}, }
• (2026). Event-triggered prescribed performance control of the multiplayer game nonlinear system via integral reinforcement learning. AMC, 511, 129716. (www)

  @article{ , abstract = {With a view to addressing the optimal control problem of multiplayer game nonlinear systems, an event-triggered prescribed performance control method based on the fusion of integral reinforcement learning (IRL) and adaptive dynamic programming (ADP) is proposed. Firstly, an auxiliary prescribed performance function (PPF) is designed to transform the original system into an unconstrained one. Drawing on the concepts of game theory, the multi-input optimal control problem is reformulated as a mixed zero-sum (MZS) game problem. Subsequently, an IRL-based event-triggered control (ETC) method is designed with a triggering condition. In this event-triggered method, ETC is updated only when the event-triggering condition is met, which reduces unnecessary communication overhead. On the basis of IRL, a critic-only neural network (NN) is established to approximate solutions of the event-triggered Hamilton-Jacobi-Bellman (HJB) equations without using the dynamic knowledge of the system. Additionally, the Lyapunov stability theorem is employed to ensure the uniform ultimate boundedness (UUB) of the system state and neural network weights. And the Zeno behavior can be avoided. Finally, an example is provided to verify the effectiveness of the proposed method in this paper.}, archive = {J_AMC}, author = {Yuanyang Hu and Jiaqi Chen and Chunbin Qin}, doi = {10.1016/j.amc.2025.129716}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129716}, shortjournal = {Appl. Math. Comput.}, title = {Event-triggered prescribed performance control of the multiplayer game nonlinear system via integral reinforcement learning}, volume = {511}, year = {2026}, }
• (2026). Numerical approximation for a stochastic time-fractional cable equation. AMC, 511, 129709. (www)

  @article{ , abstract = {An efficient numerical method is proposed to address a stochastic time-fractional cable equation driven by fractionally integrated additive noise. Under the reasonable assumptions, we rigorously establish for the first time, the existence, uniqueness, and regularity of the mild solution for this equation. For spatial discretization, a semi-discrete scheme is constructed employing the Galerkin FEM, and the optimal spatial error estimate is derived based on the semigroup approach. In temporal discretization, a piecewise constant function is introduced to approximate the noise, leading to the formulation of a regularized stochastic time-fractional cable equation. A detailed proof of the temporal error estimates is provided via the semigroup approach. Numerical experiments demonstrate that the temporal convergence order attains O ( τ 1 / 2 ) for initial data of either smooth or non-smooth type. The order is independent of the parameters α 1 ∈ ( 0 , 1 ) , α 2 ∈ ( 0 , 1 ) , and β ∈ ( 0 , 1 ) in the equation. These results perfectly align with the theoretical predictions.}, archive = {J_AMC}, author = {Qimin Li and Yubin Yan and Leijie Qiao and Yu Zhang}, doi = {10.1016/j.amc.2025.129709}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129709}, shortjournal = {Appl. Math. Comput.}, title = {Numerical approximation for a stochastic time-fractional cable equation}, volume = {511}, year = {2026}, }
• (2026). Neuro-adaptive optimized control for stochastic systems with state constraints: The non-affine faults case. AMC, 510, 129720. (www)

  @article{ , abstract = {This paper addresses the issue of neuro-adaptive optimal control for stochastic nonlinear systems with state constraints under the presence of non-affine faults. The presence of non-affine faults poses considerable challenges to the stability and performance of the system. To address these challenges, in this article, an adaptive neural network (NN) control scheme is devised within the identifier-critic-actor architecture, enabling the approximation of unknown dynamics and the design of both virtual and actual optimal controllers. In addition, Barrier Lyapunov Functions (BLFs) are utilized to ensure stability while handling state constraints, and a Butterworth low-pass filter is introduced to compensate for high-frequency noise and non-affine nonlinear faults, enhancing system robustness. Furthermore, a novel hybrid event-triggered control (HETC) strategy is proposed to reduce communication and computation demands while optimizing resource utilization. The suggested control strategy guarantees the boundedness of closed-loop signals and keeps all state variables within predefined compact sets. Lastly, the efficacy of the proposed optimal control method is demonstrated through simulation results.}, archive = {J_AMC}, author = {Tong Zhang and Yiyan Han and Ling Wang and Xin Wang}, doi = {10.1016/j.amc.2025.129720}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129720}, shortjournal = {Appl. Math. Comput.}, title = {Neuro-adaptive optimized control for stochastic systems with state constraints: The non-affine faults case}, volume = {510}, year = {2026}, }
• (2026). An entropy stable and well-balanced scheme for an augmented blood flow model with variable geometrical and mechanical properties. AMC, 510, 129719. (www)

  @article{ , abstract = {The flow of blood through a vessel can be described by a hyperbolic system of balance equations for the cross-sectional area and averaged velocity as functions of axial spatial position and time. The variable arterial wall rigidity and the equilibrium cross-sectional area are incorporated within the so-called tube law that gives rise to an internal pressure term. This system can be written as a conservative hyperbolic system for five unknowns. An entropy stable scheme for this augmented one-dimensional blood flow model is developed based on entropy conservative numerical flux. It is proved that the proposed scheme is well-balanced in the sense that it preserves both trivial (zero velocity) and non-trivial (non-zero velocity) steady-state solutions. Several demanding numerical tests show that the scheme can handle various kinds of shocks and preserves stationary solutions when geometrical and mechanical properties of the vessel are variable.}, archive = {J_AMC}, author = {Raimund Bürger and Andrés Guerra and Carlos A. Vega}, doi = {10.1016/j.amc.2025.129719}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129719}, shortjournal = {Appl. Math. Comput.}, title = {An entropy stable and well-balanced scheme for an augmented blood flow model with variable geometrical and mechanical properties}, volume = {510}, year = {2026}, }
• (2026). Switching event-triggered control for mean-square exponential synchronization of complex-valued delayed memristive neural networks under hybrid attacks. AMC, 510, 129715. (www)

  @article{ , abstract = {This paper investigates the mean-square exponential synchronization (MSES) of complex-valued delayed memristive neural networks (CVDMNNs) under complex-valued hybrid attacks. Considering the discontinuous right-hand sides of CVDMNNs, the systems are analyzed based on Filippov differential inclusion theory, and the interval matrix method is used to handle the memristive connection weights, thereby constructing error systems that are amenable to analysis. To utilize limited communication resources, a complex-valued switching event-triggered mechanism is introduced. Furthermore, to mitigate the effects of complex-valued hybrid attacks, consisting of complex-valued replay attacks and deception attacks following the Bernoulli distribution, a secure controller is designed that can simultaneously counteract the interference caused by residual terms in the error systems. On this basis, a piecewise Lyapunov functional is constructed, and the sufficient condition for achieving MSES under complex-valued hybrid attacks is derived using Lyapunov stability theory and inequality techniques. Based on the derived synchronization criterion, algorithms are proposed to determine the maximum allowable replay and deception attack rates of the system. Finally, a numerical example is provided to validate the effectiveness and superiority of the proposed scheme.}, archive = {J_AMC}, author = {Shuai Xiao and Zhen Wang and Hongjue Wang and Yueying Liu and Xia Huang}, doi = {10.1016/j.amc.2025.129715}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129715}, shortjournal = {Appl. Math. Comput.}, title = {Switching event-triggered control for mean-square exponential synchronization of complex-valued delayed memristive neural networks under hybrid attacks}, volume = {510}, year = {2026}, }
• (2026). Research on the resilience of reputation mechanism to the cooperative environment in the face of external shocks. AMC, 510, 129713. (www)

  @article{ , abstract = {Social development is changing rapidly, and the assumed stability of the cooperative environment is only an ideal state. In the majority of prior research concerning evolutionary games, it is commonly posited that external influences are absent, thereby allowing for an exclusive examination of the system’s intrinsic evolution. However, the situation in the real world is complex, dynamic, and unstable. Therefore, this paper proposes introducing an external shock that compels a change in the rate of cooperation during the evolutionary process. This approach aims to simulate environmental changes in the real world and to observe how these changes impact the rate of cooperation. When facing external shocks, the reputation mechanism has a certain degree of recovery ability, which provides certain support for the reputation theory. This paper also adds the mechanism of learning from historical strategies to the traditional reputation model, optimizing the resilience of the reputation mechanism in the face of external shocks. Simulation results show that history learning can promote the recovery of a cooperative environment faster and more stably, short memory length strengthens the role of temptation to defect, long memory amplifies the impact of the initial state, and the joint effect of history learning rate and memory length is complex.}, archive = {J_AMC}, author = {Jiaoyuan Wang and Yanlong Yang}, doi = {10.1016/j.amc.2025.129713}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129713}, shortjournal = {Appl. Math. Comput.}, title = {Research on the resilience of reputation mechanism to the cooperative environment in the face of external shocks}, volume = {510}, year = {2026}, }
• (2026). Evolutionary multigame with perception competition between conformists and imitators. AMC, 510, 129712. (www)

  @article{ , abstract = {Perceptual heterogeneity reflects individuals’ different perspectives on the same thing, and competition among individuals with different perceptions of either social dilemmas or payoffs can independently promote cooperation. In this article, we construct a coupled competition mechanism to explore the impact of competition between the two types of perceptions. Specifically, individuals’ differing perceptions of social dilemmas are modeled by the snowdrift game ( S D G ) and the prisoner’s dilemma game ( P D G ), differing perceptions of payoffs are captured by conformists and imitators. Extensive simulations show that the coupled competition mechanism markedly promotes cooperation. P D G players are substantially less competitive than S D G players, which dominate the population. Conformists act as an amplifier, inclined to choose cooperation when cooperators are dominant, thus further promoting cooperation levels. When defectors are dominant, conformists may temporarily choose to defect in the short term, but defection is unsustainable due to lower average accumulated payoffs. We further find that intensified competition among imitators weakens the competitiveness of conformists and diminishes their positive impact on cooperation. Finally, we demonstrate the impact of different proportions of conformists, network structures, and noise levels on the main results to verify the robustness of the model.}, archive = {J_AMC}, author = {Tianbo An and Ran Zhang and Jian Zhao and Huizhen Zhang and Zhen Wang}, doi = {10.1016/j.amc.2025.129712}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129712}, shortjournal = {Appl. Math. Comput.}, title = {Evolutionary multigame with perception competition between conformists and imitators}, volume = {510}, year = {2026}, }
• (2026). Prescribed-time reinforcement learning formation control of nonlinear MASs with an unknown dynamic leader. AMC, 510, 129711. (www)

  @article{ , abstract = {This paper investigates the prescribed-time optimal formation control problem for nonlinear multi-agent systems (MASs) with an unknown dynamic leader. The agents need to not only form a predefined formation pattern but also track the leader’s trajectory in a prescribed time. A hierarchical control framework, which includes the communication layer and the tracking control layer, is established to address the formulated issue. In the communication layer, a distributed prescribed-time observer is established to accurately estimate the leader’s information, which can make observation error convergences to zero in a prescribed time. In particular, the leaders’ uncertainties are solved by constructing a novel adaptive law. With the estimated results, a novel transformation relationship and the prescribed-time adjustment function are constructed to guarantee that formation tracking error converges to the predefined accuracy in a prescribed time. Subsequently, the reinforcement learning (RL) algorithm with the fuzzy logic systems (FLSs) is devised to optimize system performance. Based on the Lyapunov stability theory, it is shown that the formation errors ξ i , 1 = x i , 1 − y ^ 0 , i − η i , d are consistently confined within an interval ( tan ( − π 2 μ 2 ) , tan ( π 2 μ 2 ) ) , while all signals of the closed-loop system are bounded. Ultimately, the superiority of the devised algorithm is demonstrated by a representative example.}, archive = {J_AMC}, author = {Benxin Zhao and Yuan-Xin Li and Zhongsheng Hou}, doi = {10.1016/j.amc.2025.129711}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129711}, shortjournal = {Appl. Math. Comput.}, title = {Prescribed-time reinforcement learning formation control of nonlinear MASs with an unknown dynamic leader}, volume = {510}, year = {2026}, }
• (2026). Visualization of escher-like hyperbolic tessellations. AMC, 510, 129710. (www)

  @article{ , abstract = {By combining mathematical principles, modern computer graphics techniques, and the efforts of mathematically inclined artists, we present a visualization method for generating aesthetic patterns in hyperbolic space. To this end, we first establish fast algorithms to construct hyperbolic tilings in the Poincaré disk model. Then, using templates designed by graphic artists, we specify computer techniques to render hyperbolic tilings, which results in Escher-like patterns. Moreover, we present a simple method to realize novel hyperbolic kaleidoscopic effect. To obtain more diverse patterns, we introduce several conformal mappings to create visually appealing tessellations on the other spaces. The proposed methods can be easily implemented using shaders to obtain high-quality tessellations, which have good potential for application in the field of artistic decoration.}, archive = {J_AMC}, author = {Peichang Ouyang and Kwok Wai Chung and Alain Nicolas and Robert W. Fathauer and David Bailey and Krzysztof Gdawiec}, doi = {10.1016/j.amc.2025.129710}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129710}, shortjournal = {Appl. Math. Comput.}, title = {Visualization of escher-like hyperbolic tessellations}, volume = {510}, year = {2026}, }
• (2026). Trapezoid and trapezoidal prism for the maximum relative drawdown: Probability, crash options pricing, and risk. AMC, 510, 129708. (www)

  @article{ , abstract = {Maximum drawdown (MDD) and maximum relative drawdown (MrDD) are well-known in portfolio management and performance evaluations. They can also form the basis of a stop-loss strategy. But there is no closed-form formula for the probability that the MrDD (MDD) ever reaches some positive threshold over a period of time. This paper focuses on MrDD and employs a random walk to approximate the underlying geometric Brownian motion (GBM) for the price, taking care to match the threshold for faster convergence. Let n be the number of time steps. This paper proposes an O ( n 1.5 ) -sized trapezoid to calculate the above-mentioned probability. The trapezoid can price the crash option with a digital payoff accurately. This paper further proposes an O ( n 2.5 ) -sized trapezoidal prism to price the crash option with a resetting payoff accurately and calculate the expected return rate and common risk measures of an MrDD-based stop-loss strategy.}, archive = {J_AMC}, author = {Jia-Hao Syu and Yuh-Dauh Lyuu}, doi = {10.1016/j.amc.2025.129708}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129708}, shortjournal = {Appl. Math. Comput.}, title = {Trapezoid and trapezoidal prism for the maximum relative drawdown: Probability, crash options pricing, and risk}, volume = {510}, year = {2026}, }
• (2026). The chromatic number of {P2∪P3,banner}-free graphs. AMC, 510, 129707. (www)

  @article{ , abstract = {Borodin and Kostochka conjectured that for any graph G with maximum degree Δ ( G ) ≥ 9 , the chromatic number satisfies χ ( G ) ≤ max { Δ ( G ) − 1 , ω ( G ) } , where ω ( G ) denotes the clique number. While this conjecture remains open for general graphs, we prove its validity for the class of { P 2 ∪ P 3 , banner } -free graphs. Here, P 2 ∪ P 3 represents the disjoint union of a two-vertex path and a three-vertex path, and a banner refers to the graph formed by attaching a pendant vertex to a cycle with four vertices. Our result extends the recent work of Lan and Lin [1], establishing the conjecture for a strictly larger class of graphs.}, archive = {J_AMC}, author = {Jiali Long and Kaiyang Lan and Yan Wang}, doi = {10.1016/j.amc.2025.129707}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129707}, shortjournal = {Appl. Math. Comput.}, title = {The chromatic number of {P2∪P3,banner}-free graphs}, volume = {510}, year = {2026}, }
• (2026). Learning a robust shape parameter for RBF approximation. AMC, 510, 129706. (www)

  @article{ , abstract = {Radial basis functions (RBFs) play an important role in function interpolation, in particular, when considering an arbitrary set of interpolation nodes. The accuracy of the interpolation depends on a parameter called the shape parameter . There are many approaches in the literature on how to appropriately choose it to increase the accuracy of interpolation while avoiding stability issues. However, finding the optimal shape parameter remains a challenge in general. We introduce a new method for determining the shape parameter in RBFs. First, we construct an optimization problem to obtain a shape parameter that leads to a bounded condition number for the interpolation matrix, then, we introduce a data-driven method that controls the condition number of the interpolation matrix to avoid numerically unstable interpolations, while keeping good accuracy. In addition, a fallback procedure is proposed to enforce a strict upper bound on the condition number, as well as a learning strategy to improve the performance of the data-driven method by learning from previously run simulations. Several numerical results are presented to demonstrate the robustness of our strategy in both 1- and 2-dimensional spaces.}, archive = {J_AMC}, author = {Maria Han Veiga and Faezeh Nassajian Mojarrad and Fatemeh Nassajian Mojarrad}, doi = {10.1016/j.amc.2025.129706}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129706}, shortjournal = {Appl. Math. Comput.}, title = {Learning a robust shape parameter for RBF approximation}, volume = {510}, year = {2026}, }
• (2026). Gravitational least squares twin support vector machine based on optimal angle for class imbalance learning. AMC, 510, 129705. (www)

  @article{ , abstract = {This paper introduces the Gravitational Least Squares Twin Support Vector Machine for Class Imbalance Learning (GLSTSVM-CIL), a novel binary classification method designed to address critical limitations in existing approaches for imbalanced large-scale datasets. Traditional methods like Fuzzy TSVM and KNN-based weighting fail to simultaneously capture both global positional relationships and local density characteristics of data points. Our proposed gravitational weighting function innovatively models data samples as masses influenced by their distance from class centroids and neighborhood density, effectively prioritizing representative points while suppressing outliers. The optimization framework uniquely incorporates angular constraints between hyperplanes to enhance structural risk control and generalization capability. For scalability, we reformulate the solution into a linear system solvable via conjugate gradient methods, avoiding computationally expensive matrix inversions. Comprehensive evaluations on 92 datasets (including synthetic, noisy, medical, text, and large-scale NDC benchmarks) demonstrate GLSTSVM-CIL’s superior performance, particularly in minority-class recognition where it achieves average F1-Score improvements over baseline methods. The model maintains robust Accuracy under high noise (20 %) and extreme class imbalance (ratio 20:1) while ables to process datasets up to 50,000 samples.}, archive = {J_AMC}, author = {Abdullah Mohammadi and Jalal A. Nasiri and Sohrab Effati}, doi = {10.1016/j.amc.2025.129705}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129705}, shortjournal = {Appl. Math. Comput.}, title = {Gravitational least squares twin support vector machine based on optimal angle for class imbalance learning}, volume = {510}, year = {2026}, }
• (2026). Convergence of a positivity preserving logarithmic truncated EM method for SDEs with discontinuous drift coefficients. AMC, 510, 129704. (www)

  @article{ , abstract = {In this paper, a class of nonlinear stochastic differential equations with positive solutions and discontinuous drift coefficients is studied, considering both theoretical and computational aspects. The theoretical results focus on the existence of a unique positive solution for such SDEs via the approach introduced by Müller-Gronbach et al. (2022), and the computational aspect utilises the truncated Euler-Maruyama method proposed by Li et al. (2023) together with a logarithmic transformation that ensures a positive approximation to the original solution. The convergence of the numerical method is investigated, and the boundedness of the p -th moment is obtained. Finally, the proposed method is used to verify the convergence with the help of some numerical examples.}, archive = {J_AMC}, author = {Amir Haghighi}, doi = {10.1016/j.amc.2025.129704}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129704}, shortjournal = {Appl. Math. Comput.}, title = {Convergence of a positivity preserving logarithmic truncated EM method for SDEs with discontinuous drift coefficients}, volume = {510}, year = {2026}, }
• (2026). Enumerating the number of k-matchings in successively amalgamated graphs. AMC, 510, 129703. (www)

  @article{ , abstract = {In this paper, the transfer matrix technique using the k -matching vector is developed to compute the number of k -matchings in an arbitrary graph which can be constructed by successive amalgamations over sets of cardinality two. This widely extends known methods from the literature developed for computing the number of k -matchings in benzenoid chains, octagonal chains, cyclooctatetraene chains, and arbitrary cyclic chains. Two examples demonstrating how the present method can be applied are given, one of them being an elaborated chemical example.}, archive = {J_AMC}, author = {Simon Grad and Sandi Klavžar}, doi = {10.1016/j.amc.2025.129703}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129703}, shortjournal = {Appl. Math. Comput.}, title = {Enumerating the number of k-matchings in successively amalgamated graphs}, volume = {510}, year = {2026}, }
• (2026). Community detection methods for GBF-PUM signal approximation on graphs. AMC, 510, 129702. (www)

  @article{ , abstract = {Graph signal approximation plays a key role in processing irregularly distributed data on graphs, where achieving smooth and computationally efficient interpolation is essential. In this work, we introduce a new approach that combines a spectral community detection technique with the partition of unity method (PUM) applied to signal approximation on graphs. The PUM provides an effective technique for handling irregularly distributed data by dividing the graph into smaller subgraphs, constructing local interpolants and combining them to produce a global approximation. Since the first step in the PUM consists in dividing the graph into disjoint communities, we focus in particular on exploring and testing some community detection algorithms based on the maximization of the modularity. Then, we integrate the PUM with a local graph basis function approximation scheme, resulting in an accurate and computationally efficient approach for graph signal approximation.}, archive = {J_AMC}, author = {Roberto Cavoretto and Chiara Comoglio and Alessandra De Rossi}, doi = {10.1016/j.amc.2025.129702}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129702}, shortjournal = {Appl. Math. Comput.}, title = {Community detection methods for GBF-PUM signal approximation on graphs}, volume = {510}, year = {2026}, }
• (2026). Dynamics of two-strain disease transmission with negative vaccine information and vaccination on multiplex networks. AMC, 510, 129701. (www)

  @article{ , abstract = {During the disease transmission, the virus may mutate and novel strains appear. That causes the reduction of the immune effect of the vaccine against the original strain, and the vaccinated individuals may also be infected, which would lead to the generation and dissemination of negative vaccine information. However, the coupled dynamics of multi-strain disease and negative vaccine information has not yet been comprehensively investigated. In this paper, we consider both the original and mutant strains of the disease and introduce vaccination behavior in the propagation process to expand the classic susceptible-infectious-susceptible (SIS) model into five states: susceptible, vaccinated, infected by original strain, infected by mutant strain and recovered. Then, a two-layer network model is proposed to describe the coupled propagation dynamics of the negative vaccine information and the two-strain disease, and the epidemic outbreak thresholds of the two-strain disease are derived by using the Microscopic Markov Chain Approach (MMCA). Subsequently, the numerical results derived from the MMCA are compared with those from Monte-Carlo (MC) simulation to validate the correctness of the former. Finally, the phase diagrams and time history curves are presented to conduct parametric studies on the epidemic thresholds and information-disease transmission dynamics, such as the vaccination cost, vaccine effectiveness and the negative vaccine information. This research demonstrates that reducing the vaccination cost and enhancing the positive publicity of vaccine can effectively prevent the disease spreading when the infection rate of the mutant strain is low. If this infection rate is large, the effective measure is developing specific vaccines against the mutant strain to improve the vaccination efficacy. Those findings provide valuable strategies for the authorities to make appropriate disease prevention measures.}, archive = {J_AMC}, author = {Lun Liu and Shaorui Geng and Lilin Wei and Zhenyong Lu and Yinghong Ma}, doi = {10.1016/j.amc.2025.129701}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129701}, shortjournal = {Appl. Math. Comput.}, title = {Dynamics of two-strain disease transmission with negative vaccine information and vaccination on multiplex networks}, volume = {510}, year = {2026}, }
• (2026). Induced cycles vertex number and (1,2)-domination in cubic graphs. AMC, 510, 129700. (www)

  @article{ , abstract = {A (1,2)-dominating set in a graph G is a set S such that every vertex outside S has at least one neighbor in S , and each vertex in S has at least two neighbors in S . The (1,2)-domination number, γ 1 , 2 ( G ) , is the minimum size of such a set, while c i n d ( G ) is the cardinality of the largest vertex set in G that induces one or more cycles. In this paper, we initiate the study of a relationship between these two concepts and discuss how establishing such a connection can contribute to solving a conjecture on the lower bound of c i n d ( G ) for cubic graphs. We also establish an upper bound on c i n d ( G ) for cubic graphs and characterize graphs that achieve this bound.}, archive = {J_AMC}, author = {Rija Erveš and Aleksandra Tepeh}, doi = {10.1016/j.amc.2025.129700}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129700}, shortjournal = {Appl. Math. Comput.}, title = {Induced cycles vertex number and (1,2)-domination in cubic graphs}, volume = {510}, year = {2026}, }
• (2026). A c++ implementation of the discrete adjoint sensitivity analysis method for explicit adaptive runge-kutta methods enabled by automatic adjoint differentiation and SIMD vectorization. AMC, 510, 129699. (www)

  @article{ , abstract = {A C++ library for sensitivity analysis of optimisation problems involving ordinary differential equations (ODEs) enabled by automatic differentiation (AD) and SIMD (Single Instruction, Multiple data) vectorization is presented. The discrete adjoint sensitivity analysis method is implemented for adaptive explicit Runge-Kutta (ERK) methods. Automatic adjoint differentiation (AAD) is employed for efficient evaluations of products of vectors and the Jacobian matrix of the right hand side of the ODE system. This approach avoids the low-level drawbacks of the black box approach of employing AAD on the entire ODE solver and opens the possibility to leverage parallelization. SIMD vectorization is employed to compute the vector-Jacobian products concurrently. We study the performance of other methods and implementations of sensitivity analysis and we find that our algorithm presents a small advantage compared to equivalent existing software.}, archive = {J_AMC}, author = {Rui Martins and Evgeny Lakshtanov}, doi = {10.1016/j.amc.2025.129699}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129699}, shortjournal = {Appl. Math. Comput.}, title = {A c++ implementation of the discrete adjoint sensitivity analysis method for explicit adaptive runge-kutta methods enabled by automatic adjoint differentiation and SIMD vectorization}, volume = {510}, year = {2026}, }
• (2026). Components of flip graphs of domino tilings in quadriculated cylinder and torus. AMC, 510, 129697. (www)

  @article{ , abstract = {Let R be a quadriculated surface, possibly with boundary, consisting of unit squares, and each interior vertex being surrounded by 4 squares. A tiling of R is a placement of dominoes (a pair of adjacent squares) so that there are no gaps or overlaps. The flip graph of R is a graph whose vertices are all tilings of R and two tilings are adjacent if we can obtain one from another by a flip ( 90 ∘ rotation of a pair of side-by-side dominoes). By using graph-theoretical approach, we prove that the flip graph of 2 m × ( 2 n + 1 ) quadriculated cylinder is still connected, but that of 2 m × ( 2 n + 1 ) quadriculated torus consists of two isomorphic components. For a tiling t , we associate an integer, called forcing number, as the minimum number of dominoes in t that is contained in no other tilings. As a consequence, we obtain that the forcing numbers of all tilings in 2 m × ( 2 n + 1 ) quadriculated cylinder and torus form respectively an integer interval whose maximum value is ( n + 1 ) m .}, archive = {J_AMC}, author = {Qianqian Liu and Jingfeng Wang and Chunmei Li and Heping Zhang}, doi = {10.1016/j.amc.2025.129697}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129697}, shortjournal = {Appl. Math. Comput.}, title = {Components of flip graphs of domino tilings in quadriculated cylinder and torus}, volume = {510}, year = {2026}, }
• (2026). An equitable partition based construction of graphs with the same spectral radius. AMC, 510, 129696. (www)

  @article{ , abstract = {An equitable partition of a graph is a remarkable tool that provides valuable spectral information. Among other properties, it is known that the spectral radius of the divisor matrix of any equitable partition equals the spectral radius of the graph. Recently, several methods for construction of families of graphs with the same spectral radius emerged in the literature. In this paper, we develop a unified approach to cyclic-graph constructions by formulating a general method for generating families of connected graphs with a common spectral radius, grounded in the theory of equitable partitions.}, archive = {J_AMC}, author = {Milica Anđelić and Carlos M. da Fonseca and Zoran Stanić}, doi = {10.1016/j.amc.2025.129696}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129696}, shortjournal = {Appl. Math. Comput.}, title = {An equitable partition based construction of graphs with the same spectral radius}, volume = {510}, year = {2026}, }
• (2026). On reduced hamilton walks. AMC, 510, 129695. (www)

  @article{ , abstract = {A Hamilton walk in a finite graph is a walk, either open or closed, that traverses every vertex at least once. Here, we introduce Hamilton walks that are reduced in the sense that they avoid immediate backtracking: a reduced Hamilton walk never traverses the same edge forth and back consecutively. While every connected graph admits a Hamilton walk, existence of a reduced Hamilton walk is not guaranteed for all graphs. However, we prove that a reduced Hamilton walk does exist in a connected graph with minimal valency at least 2. Furthermore, given such a graph on n vertices, we present an O ( n 2 ) -time algorithm that constructs a reduced Hamilton walk of length at most n ( n + 3 ) / 2 . Specifically, for a graph belonging to a family of regular expander graphs, we can find a reduced Hamilton walk of length at most c ( 6 n − 2 ) log n + 2 n , where c is a constant independent of n .}, archive = {J_AMC}, author = {Aleksander Malnič and Rok Požar}, doi = {10.1016/j.amc.2025.129695}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129695}, shortjournal = {Appl. Math. Comput.}, title = {On reduced hamilton walks}, volume = {510}, year = {2026}, }
• (2026). Šoltés problem for the kirchhoff index of a graph. AMC, 510, 129694. (www)

  @article{ , abstract = {The Kirchhoff index K f ( G ) of a connected graph G is defined as the sum of resistance distances between all pairs of vertices in G . We say that v ∈ V ( G ) is a good vertex if the Kirchhoff index remains unchanged when v is removed, i.e. K f ( G ) = K f ( G − v ) . In 1991, Šoltés studied the Wiener index of a graph and posed the problem of identifying graphs for which the removal of an arbitrary vertex preserves the Wiener index. In this paper, we explore a similar concept: identifying Kirchhoff Šoltés graphs , i.e. graphs in which all vertices are good vertices. We show that the cycle C 5 is a Kirchhoff Šoltés graph. Due to the challenge of finding more examples of such graphs, we shift our focus to several relaxed versions of the Kirchhoff Šoltés problem, where the primary objective is to identify graphs containing at least one good vertex. One of them is the β - Kirchhoff Šoltés problem , which seeks to find an infinite family of graphs in which the proportion of good vertices is at least β , with β ∈ ( 0 , 1 ] being a specified rational number. Another one involves constructing infinite families of graphs where the proportion of good vertices increases and asymptotically approaches a given real number γ ∈ ( 0 , 1 ] as the order of the graph grows. We demonstrate that both relaxed versions have infinitely many solutions. In particular, we prove the existence of infinitely many graphs for which the proportion β of good vertices, 1 / 7 ≤ β < 1 / 5 tends to a certain irrational number. Furthermore, we prove the existence of infinitely many graphs with half good vertices, and for each s ∈ N , we construct an infinite family of graphs whose proportion of good vertices tends to s + 1 2 s + 1 . These findings could be pivotal in addressing the original problem of determining whether there are additional solutions beyond C 5 .}, archive = {J_AMC}, author = {Kurt Klement Gottwald and Snježana Majstorović Ergotić and Tomislav Došlić}, doi = {10.1016/j.amc.2025.129694}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129694}, shortjournal = {Appl. Math. Comput.}, title = {Šoltés problem for the kirchhoff index of a graph}, volume = {510}, year = {2026}, }
• (2026). Secure observer-driven synchronization for complex networks constrained by partial information exchange and limited communication bandwidth. AMC, 510, 129693. (www)

  @article{ , abstract = {This paper addresses the observer-driven secure synchronization control issue for partial information-based complex networks (CNs) subject to bandwidth constraints. Firstly, given that the complete data transmission between CN nodes can not be always realized in practice, the system model of CNs with partial information exchange is specifically introduced. Then, considering the limitation of communication bandwidth, weighted try-once-discard (WTOD) protocol is utilized to enhance the signal transmission efficiency within each node. The delivery of the data adhered to WTOD protocol is further assumed to be compromised by false-data-injection (FDI) attack, which is a well-recognized risk in communication networks. Afterward, distributed secure observer-based synchronization controllers are designed accordingly. Based on defining the observation and synchronization errors, an augmented system model is then established, and the gain matrices for observers and controllers are obtained through an analysis of the stability conditions of the augmented system. Lastly, a numerical example is provided to validate the proposed method’s effectiveness.}, archive = {J_AMC}, author = {Yan Li and Qirun Hang and Lijuan Zha and Jinliang Liu}, doi = {10.1016/j.amc.2025.129693}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129693}, shortjournal = {Appl. Math. Comput.}, title = {Secure observer-driven synchronization for complex networks constrained by partial information exchange and limited communication bandwidth}, volume = {510}, year = {2026}, }
• (2026). Causal state feedback representation for infinite horizon LQ problems of fractional systems. AMC, 510, 129692. (www)

  @article{ , abstract = {In this paper, we consider the feedback representation problem of the optimal control for infinite horizon linear quadratic optimal control problems of fractional systems. By employing the characterization of the optimal control pair and a family of projection operators, we derive the causal state feedback representations which depend only on the history value of the state and does not rely on further state value. Finally, two illustrative examples are included.}, archive = {J_AMC}, author = {Jianping Huang and Huacheng Zhou}, doi = {10.1016/j.amc.2025.129692}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129692}, shortjournal = {Appl. Math. Comput.}, title = {Causal state feedback representation for infinite horizon LQ problems of fractional systems}, volume = {510}, year = {2026}, }
• (2026). Observer-based leader-follower bipartite consensus for linear multiagent systems with communication link faults. AMC, 510, 129690. (www)

  @article{ , abstract = {This article investigates the observer-based leader-follower bipartite consensus (LFBC) problem for linear multiagent systems (MASs) with unknown communication link faults. Firstly, by designing the corresponding leader state observers for the leader and each follower respectively, and adaptively adjusting the coupling strength, the weight changes due to unknown communication link faults are solved. Secondly, novel double event-triggered mechanisms (DETMs) based on observers are proposed, which can realize communication and observers updating asynchronously and help to save communication resources. The triggering threshold is designed based on the adaptive leader state error, avoiding the need for continuous information exchange with neighboring agents. Afterwards, our proposed leader state observer-based approaches are applied to the control, aiming to ensure system performance and stabilize the handling of system faults. Finally, two simulation examples are presented to verify the effectiveness of the proposed control strategy.}, archive = {J_AMC}, author = {Dongsheng Yang and Jing He and Jiayue Sun and Juan Zhang}, doi = {10.1016/j.amc.2025.129690}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129690}, shortjournal = {Appl. Math. Comput.}, title = {Observer-based leader-follower bipartite consensus for linear multiagent systems with communication link faults}, volume = {510}, year = {2026}, }
• (2026). Bipartite containment control of multi-agent systems under channel imperfections: When bit-flips meet packet-losses. AMC, 510, 129689. (www)

  @article{ , abstract = {This paper investigates bipartite containment control of linear multi-agent systems under channel imperfections using source binary coding schemes with bit-flips and packet-losses. The communication channels are allocated through an average allocation protocol. The source binary coding scheme for measurement outputs is employed, and a finite-length bit stream is designed to satisfy the requirement of channel capacity. The paper aims to establish mean-square bipartite containment conditions under communication constraints such that the containment error is bounded. First, the statistical properties of the aggregated error are derived. Then, a sufficient condition is obtained to ensure bipartite containment in the mean-square sense. Furthermore, with the free matrix approach and the matrix inequality method, the control gain is obtained to ensure the existence of a desired controller. Finally, two examples are provided to illustrate the effectiveness of the results.}, archive = {J_AMC}, author = {Shaobo Zheng and Lei Zhou}, doi = {10.1016/j.amc.2025.129689}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129689}, shortjournal = {Appl. Math. Comput.}, title = {Bipartite containment control of multi-agent systems under channel imperfections: When bit-flips meet packet-losses}, volume = {510}, year = {2026}, }
• (2026). Asymptotic analysis of the first eigenvalue for the sturm-liouville problems with coupled boundary conditions. AMC, 510, 129688. (www)

  @article{ , abstract = {The present paper is concerned with the asymptotic behavior of the first eigenvalue on the jump set of the Sturm-Liouville eigenvalue problems with coupled boundary conditions. By transforming the original problem into two problems with separated boundary conditions, we obtain the asymptotic formula of the first eigenvalue and the first normalized eigenfunction in terms of the coefficients in the equations. Particularly, we prove that the unique zero of the first eigenfunction tends to the zero of the first eigenfunction of the corresponding Laplace problem. The results indicate that the asymptotic behavior is only related to the values of the coefficients in the neighborhood of the endpoints.}, archive = {J_AMC}, author = {Xinyu Zhang and Jiangang Qi and Chunyan Sun}, doi = {10.1016/j.amc.2025.129688}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129688}, shortjournal = {Appl. Math. Comput.}, title = {Asymptotic analysis of the first eigenvalue for the sturm-liouville problems with coupled boundary conditions}, volume = {510}, year = {2026}, }
• (2026). Stokes–Brinkman–Darcy models for fluid–porous systems: Derivation, analysis and validation. AMC, 510, 129687. (www)

  @article{ , abstract = {Flow interaction between a plain-fluid region in contact with a porous layer attracted significant attention from modelling and analysis sides due to numerous applications in biology, environment and industry. In the most widely used coupled model, fluid flow is described by the Stokes equations in the free-flow domain and Darcy’s law in the porous medium, and complemented by the appropriate interface conditions. However, traditional coupling concepts are restricted, with a few exceptions, to one-dimensional flows parallel to the fluid–porous interface. In this work, we use an alternative approach to model interaction between the plain-fluid domain and porous medium by considering a transition zone, and propose the full- and hybrid-dimensional Stokes–Brinkman–Darcy models. In the first case, the equi-dimensional Brinkman equations are considered in the transition region, and the appropriate interface conditions are set on the top and bottom of the transition zone. In the latter case, we perform a dimensional model reduction by averaging the Brinkman equations in the normal direction and using the proposed transmission conditions. The well-posedness of both coupled problems is proved, and some numerical simulations are carried out in order to validate the concepts.}, archive = {J_AMC}, author = {Linheng Ruan and Iryna Rybak}, doi = {10.1016/j.amc.2025.129687}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129687}, shortjournal = {Appl. Math. Comput.}, title = {Stokes–Brinkman–Darcy models for fluid–porous systems: Derivation, analysis and validation}, volume = {510}, year = {2026}, }
• (2026). Dynamics of cooperative evolution with leader-follower hierarchy under information uncertainty in two-layer grid networks. AMC, 510, 129686. (www)

  @article{ , abstract = {Drawing on the idea of leader-follower game theory, this paper investigates the evolution of cooperative behavior in two-layer grid networks. The two-layer network is divided into a leader layer (upper layer) and a follower layer (lower layer), where the snowdrift game (SDG) and the prisoner’s dilemma game (PDG) are conducted, respectively. Based on an updating mechanism of the modified Fermi probability function that incorporates strategy alignment, the leader layer first completes strategy update, and the follower layer then updates its strategy after acquiring information from the upper layer via perception probability. Numerical simulations reveal that the cost-benefit ratio significantly influences the differences in perception strategies between layers. Specifically, at lower ratios, there is a positive correlation between followers’ cooperation and their perception probability, whereas leaders exhibit a negative correlation. The opposite trends are observed at higher ratios. Compared with single-layer networks, the leader-follower structure with two-layer networks significantly enhances the cooperation in the follower layer and also lightly fosters the cooperation in leader layer. Additionally, factors such as inter-layer coupling strength, rewards and punishments for strategy consistency, and selection intensity all promote the emergence of cooperation to varying degrees. This leader-follower hierarchical game offers a new research paradigm for interaction mechanisms in complex social systems.}, archive = {J_AMC}, author = {Bolin Yang and Guanghui Yang}, doi = {10.1016/j.amc.2025.129686}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129686}, shortjournal = {Appl. Math. Comput.}, title = {Dynamics of cooperative evolution with leader-follower hierarchy under information uncertainty in two-layer grid networks}, volume = {510}, year = {2026}, }
• (2026). Reinforcement learning in evolutionary game theory: A brief review of recent developments. AMC, 510, 129685. (www)

  @article{ , abstract = {With the rapid progress of artificial intelligence, the integration of evolutionary game theory and reinforcement learning has become a hot research frontier in the last years. Evolutionary game theory provides a mathematical framework for depicting the strategy interaction among individuals, traditionally based on pre-defined, rule-based strategy update protocols. In contrast, reinforcement learning enables agents to adaptively select optimal actions through trial-and-error learning, hence better reflecting real-world decision-making. These complementary features create the foundation for their convergence. Our paper presents a didactic review of contemporary reinforcement learning applications in evolutionary game theory, focusing on those recently published works which open novel research paths to enrich our understanding of mutualistic cooperation. We summarize major concepts and terms, including the basic problem of collective cooperation, modeling of complex population dynamics, influence of algorithmic parameters, and the combination of deep learning. Finally, we discuss prospects for this interdisciplinary field, emphasizing the importance of intelligent learning through the lens of evolutionary game.}, archive = {J_AMC}, author = {Kai Xie and Attila Szolnoki}, doi = {10.1016/j.amc.2025.129685}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129685}, shortjournal = {Appl. Math. Comput.}, title = {Reinforcement learning in evolutionary game theory: A brief review of recent developments}, volume = {510}, year = {2026}, }
• (2026). Observer-based reachable set synthesis for multi-agent systems: A sampling-based distributed event-triggered strategy. AMC, 510, 129683. (www)

  @article{ , abstract = {This paper investigates the problem of reachable set synthesis for a class of multi-agent systems. Considering the limited resources for communication between agents, a dual-end sampling mechanism is introduced to reduce the communication load. For the sampled signals at the output end, the cubic spline interpolation method is used to construct a fitting function to approximate the dynamic output signals, and a state observer is constructed according to the fitting function to estimate the system state of each agent. At the input side, the sampling distributed event triggering mechanism is used to design an event-based distributed observation control strategy to ensure that the tracking error is bounded. Finally, the results of the simulations substantiate the validity of the conclusions.}, archive = {J_AMC}, author = {Mingde Liu and Liang Zhang and Ning Zhao and Xinjun Wang and Yongchao Liu}, doi = {10.1016/j.amc.2025.129683}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129683}, shortjournal = {Appl. Math. Comput.}, title = {Observer-based reachable set synthesis for multi-agent systems: A sampling-based distributed event-triggered strategy}, volume = {510}, year = {2026}, }
• (2026). Energy-conserving kansa methods for hamiltonian wave equations. AMC, 510, 129682. (www)

  @article{ , abstract = {We introduce a fast, constrained meshfree solver designed specifically to inherit energy conservation (EC) in second-order time-dependent Hamiltonian wave equations. For discretization, we adopt the Kansa method, also known as the kernel-based collocation method, combined with time-stepping. This approach ensures that the critical structural feature of energy conservation is maintained over time by embedding a quadratic constraint into the definition of the numerical solution. To address the computational challenges posed by the nonlinearity in the Hamiltonian wave equations and the EC constraint, we propose a fast iterative solver based on the Newton method with successive linearization. This novel solver significantly accelerates the computation, making the method highly effective for practical applications. Numerical comparisons with the traditional secant methods highlight the competitive performance of our scheme. These results demonstrate that our method not only conserves the energy but also offers a promising new direction for solving Hamiltonian wave equations more efficiently. While we focus on the Kansa method and corresponding convergence theories in this study, the proposed solver is based solely on linear algebra techniques and has the potential to be applied to EC constrained optimization problems arising from other PDE discretization methods.}, archive = {J_AMC}, author = {Xiaobin Li and Meng Chen and Zhengjie Sun and Leevan Ling and Siqing Li}, doi = {10.1016/j.amc.2025.129682}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129682}, shortjournal = {Appl. Math. Comput.}, title = {Energy-conserving kansa methods for hamiltonian wave equations}, volume = {510}, year = {2026}, }
• (2026). High order nonstandard finite-difference methods. AMC, 510, 129681. (www)

  @article{ , abstract = {Nonstandard finite difference (NSFD) methods have been considered to overcome some issues of standard methods, particularly when the numerical solution must preserve important properties of the exact solution. These issues increase for high order methods. In this paper we first derive a general procedure to obtain unconditionally positive second order NSFD methods. Furthermore, by suitably adding some parameters α i within these schemes, we show that it is still possible to get positivity, and also to preserve other qualitative properties of the exact solution. In fact, for each particular problem we can get optimal values of α i that guarantee positivity, elementary stability and the minimization of the local truncation error, being possible to achieve also third order nonstandard schemes, which are not present in the literature. As an example of use, we employ the developed theory to derive positive and elementary stable NSFD methods of order one, two and three for a predator-prey model, showing their advantages over other nonstandard methods from the literature.}, archive = {J_AMC}, author = {D. Conte and G. Pagano and T. Roldán}, doi = {10.1016/j.amc.2025.129681}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129681}, shortjournal = {Appl. Math. Comput.}, title = {High order nonstandard finite-difference methods}, volume = {510}, year = {2026}, }
• (2026). Consensus control for multi-agent systems with unknown fading channels and limited bandwidth. AMC, 510, 129680. (www)

  @article{ , abstract = {This paper is concerned with the consensus of multi-agent systems (MASs) in the presence of fading channels and limited communication bandwidth. These issues will lead to inaccurate and energy constraints of communication data respectively, which will not be possible for MASs to achieve consensus. To this end, this paper proposes a variable structure encoding-decoding scheme (VS-EDS), which includes an encoder with a unitary signal generator and a corresponding decoder with an adaptive activation module. The advantages of the proposed VS-EDS are mainly in three aspects. First, it avoids the need for existing methods to perform a large amount of statistical work on the statistics of fading gains before the system runs. Second, it can eliminate the communication errors caused by fading. At the same time, it can adapt to the changing communication network environment. In addition, in order to achieve consensus control, a distributed protocol based on the VS-EDS is designed. Finally, the feasibility and effectiveness of the VS-EDS are demonstrated through comparative simulations.}, archive = {J_AMC}, author = {Caibin Yao and Yiwen Qi and Ziyu Qu and Wu Wang and Zonghua Zheng}, doi = {10.1016/j.amc.2025.129680}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129680}, shortjournal = {Appl. Math. Comput.}, title = {Consensus control for multi-agent systems with unknown fading channels and limited bandwidth}, volume = {510}, year = {2026}, }
• (2026). On deep learning for computing the dynamic initial margin and margin value adjustment. AMC, 510, 129679. (www)

  @article{ , abstract = {The present work addresses the challenge of training neural networks for Dynamic Initial Margin (DIM) computation in counterparty credit risk, a task traditionally burdened by the high costs associated with generating training datasets through nested Monte Carlo (MC) simulations. By condensing the initial market state variables into an input vector, determined through an interest rate model and a parsimonious parameterization of the current interest rate term structure, we construct a training dataset where the labels are future realizations, generated with a single MC path, of the Initial Margin (IM) variable. Since DIM is defined as the conditional expectation of IM, the latter can be understood as noisy and unbiased samples of DIM, allowing the application of deep learning regression techniques to its computation. To this end, a multi-output neural network structure is employed to handle DIM as a time-dependent function, facilitating training across a mesh of monitoring times. This methodology offers significant advantages: it reduces the dataset generation cost to a single MC execution and parameterizes the neural network by initial market state variables, obviating the need for repeated training. Experimental results demonstrate the approach’s convergence properties and robustness across different interest rate models (Hull-White and Cox-Ingersoll-Ross) and portfolio complexities, validating its general applicability and efficiency in more realistic scenarios.}, archive = {J_AMC}, author = {Joel P. Villarino and Alvaro Leitao}, doi = {10.1016/j.amc.2025.129679}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129679}, shortjournal = {Appl. Math. Comput.}, title = {On deep learning for computing the dynamic initial margin and margin value adjustment}, volume = {510}, year = {2026}, }
• (2026). On solutions of a coupled system of quantum integral equations in banach spaces. AMC, 510, 129678. (www)

  @article{ , abstract = {This attempt studies the existence of solutions for a coupled system of quantum integral and quadratic integral equations via generalized Darbo’s fixed point theorem, namely, Petryshyan’s fixed point theorem. Finally, we conclude with a concrete numerical example, which confirms that our results can apply to a wide range of quantum integral equations.}, archive = {J_AMC}, author = {Hamid Reza Sahebi and Manochehr Kazemi and Mohamed M.A. Metwali}, doi = {10.1016/j.amc.2025.129678}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129678}, shortjournal = {Appl. Math. Comput.}, title = {On solutions of a coupled system of quantum integral equations in banach spaces}, volume = {510}, year = {2026}, }
• (2026). Arbitrary-order sensitivity analysis of frequency response functions using hypercomplex automatic differentiation and spectral finite elements. AMC, 510, 129677. (www)

  @article{ , abstract = {Accurately computing sensitivities of Frequency Response Functions (FRFs) is crucial for analyzing the dynamic behavior of structures by enabling quantifying the impact that variations in geometry, material properties, and boundary conditions have on their dynamic response. However, one of the primary challenges in calculating accurate sensitivities lies in the numerical differentiation required to estimate the sensitivities of the FRFs. This paper presents a new method called the Hypercomplex Spectral Finite Elements Method (HYPAD-SFEM). HYPAD-SFEM combines the HYPercomplex Automatic Differentiation method (HYPAD) with the Spectral Finite Elements Method (SFEM) to compute highly accurate arbitrary-order sensitivities of the FRFs. To demonstrate and verify the method's performance and accuracy, we analyzed a truss structure under a harmonic axial load and compared the results with analytical equations, Finite Differences (FD), and traditional Automatic Differentiation (AD). Excellent agreement was observed between the computed displacements and their sensitivities, considering material properties, geometry, and boundary conditions. The application of HYPAD-SFEM was then extended to a more complex problem by performing shape sensitivity analysis of the dynamic behavior of a phononic lattice. Again, excellent agreement was found between HYPAD, FD and AD. In general, the proposed HYPAD-SFEM ensures high accuracy independent of the perturbation step selection, alleviating FD’s fundamental issues. Moreover, HYPAD-SFEM delivers superior computational performance when compared with traditional AD. Hence, HYPAD-SFEM provides an effective approach for FRF sensitivity analysis, facilitating design optimization, parameter tuning, robustness analysis, and model updating and validation in structural dynamics.}, archive = {J_AMC}, author = {Juan David Navarro and Juan C. Velasquez-Gonzalez and Mauricio Aristizabal and Arturo Montoya and Harry R. Millwater and David Restrepo}, doi = {10.1016/j.amc.2025.129677}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129677}, shortjournal = {Appl. Math. Comput.}, title = {Arbitrary-order sensitivity analysis of frequency response functions using hypercomplex automatic differentiation and spectral finite elements}, volume = {510}, year = {2026}, }
• (2026). Hinderance of cooperation by individual solutions: Evolutionary dynamics of three-strategy games combining the prisoner’s dilemma and stag hunt. AMC, 510, 129676. (www)

  @article{ , abstract = {This study analyzes a three-strategy game that integrates elements of the prisoner’s dilemma and stag hunt. The framework builds upon recent research that determines the impact of individual solutions. Individuals adopting such solutions do not free ride on the cooperative efforts of others; rather, they act just sufficiently to prevent adverse consequences for themselves. We hypothesize that individual solutions function analogously to the defection strategy in the stag hunt game. This study examines the effects of orthodox free riding and individual solutions on the evolution of cooperation. Our analysis reveals that in well-mixed populations, the only stable rest point is the one in which all players opt for individual solutions. When interactions are structured on a square lattice, cooperation levels exhibit modest improvements, primarily sustained through cyclic dominance. Payoff values favorable to cooperation result in full cooperation; however, rare mutations can disrupt the cooperative equilibrium, promoting the adoption of individual solutions. Our analysis aligns with experimental observations and illustrates the importance of overcoming reliance on individual solutions to better understand the emergence of cooperation.}, archive = {J_AMC}, author = {Hirofumi Takesue}, doi = {10.1016/j.amc.2025.129676}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129676}, shortjournal = {Appl. Math. Comput.}, title = {Hinderance of cooperation by individual solutions: Evolutionary dynamics of three-strategy games combining the prisoner’s dilemma and stag hunt}, volume = {510}, year = {2026}, }
• (2026). Stability analysis of Runge–Kutta methods for nonlinear delay-integro-differential-algebraic equations. AMC, 510, 129675. (www)

  @article{ , abstract = {This paper is devoted to examining the stability of Runge–Kutta methods for solving nonlinear delay-integro-differential-algebraic equations (DIDAEs). The stability of exact solution for nonlinear DIDAEs is obtained by using the Halanay’s inequality. Hybrid numerical schemes combining Runge–Kutta methods and compound quadrature rules are analyzed for nonlinear DIDAEs. Criteria for ensuring the global and asymptotic stability of the proposed schemes are established. Several numerical examples are provided to validate the theoretical findings.}, archive = {J_AMC}, author = {Gehao Wang and Yuexin Yu}, doi = {10.1016/j.amc.2025.129675}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129675}, shortjournal = {Appl. Math. Comput.}, title = {Stability analysis of Runge–Kutta methods for nonlinear delay-integro-differential-algebraic equations}, volume = {510}, year = {2026}, }
• (2026). Sine-transform-based fast solvers for riesz fractional nonlinear schrödinger equations with attractive nonlinearities. AMC, 510, 129674. (www)

  @article{ , abstract = {This paper presents fast solvers for linear systems arising from the discretization of fractional nonlinear Schrödinger equations with Riesz derivatives and attractive nonlinearities. These systems exhibit complex symmetry, indefiniteness, and a d -level diagonal-plus-Toeplitz structure. We propose a Toeplitz-based anti-symmetric and normal splitting iteration method for the equivalent real block linear systems, ensuring unconditional convergence. By integrating this iteration method with sine-transform-based preconditioning, we introduce a novel preconditioner that enhances the convergence rate of Krylov subspace methods. Both theoretical and numerical analyses demonstrate that the new preconditioner exhibits a parameter-free property, and favorable eigenvalue clustering nature of the corresponding preconditioned coefficient matrix, and the associated preconditioned GMRES method converges independently of the mesh size in space and the order of Riesz fractional derivatives.}, archive = {J_AMC}, author = {Chao Chen and Xi Yang and Fei-Yan Zhang}, doi = {10.1016/j.amc.2025.129674}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129674}, shortjournal = {Appl. Math. Comput.}, title = {Sine-transform-based fast solvers for riesz fractional nonlinear schrödinger equations with attractive nonlinearities}, volume = {510}, year = {2026}, }
• (2026). A meshless point collocation solver for elliptic boundary value problems. AMC, 510, 129673. (www)

  @article{ , abstract = {We propose a strong-form meshless point collocation (MPC) solver for the Poisson equation. The Poisson equation allows us to compute approximate pressure corrections and ensure the incompressibility of the velocity field or to solve for the stream function and vorticity. We discretize the spatial domain using quadratic triangular elements in 2D and tetrahedral elements in 3D. The element nodes, including the vertices and edge midpoints, define the point cloud used in the MPC method. We determine the support domain for each using the mesh’s connectivity. When constructing the stiffness matrix, the resulting algebraic systems have the same bandwidth as those generated by the finite element (FE) method. We use direct and iterative solvers to assess the accuracy and efficiency of the MPC method. Our solution strategy enables automatic mesh generation, as nodes are utilized directly in the interpolation construction, eliminating the need to evaluate mesh quality. Finally, we investigate the efficiency of the MPC method in solving linear systems in 3D with a large number of nodes.}, archive = {J_AMC}, author = {G.C. Bourantas and A. Sakellarios and N. Malamos and V.C. Loukopoulos and V.N. Burganos and D.I. Fotiadis and V.M. Calo}, doi = {10.1016/j.amc.2025.129673}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129673}, shortjournal = {Appl. Math. Comput.}, title = {A meshless point collocation solver for elliptic boundary value problems}, volume = {510}, year = {2026}, }
• (2026). Solving fuzzy non-homogeneous linear differential systems with piecewise constant arguments involving the short-memory variable-order caputo fractional derivative. AMC, 510, 129672. (www)

  @article{ , abstract = {We establish an explicit solution formula for a linear class of fractional differential systems with piecewise constant arguments (FDSs-PCA), incorporating generalized variable-order fractional derivatives (VO-FDs) of order in the interval (0,1). The effectiveness and practical relevance of our result are demonstrated through several illustrative examples.}, archive = {J_AMC}, author = {Nguyen Dinh Phu and Lai Van Phut and Ngo Van Hoa}, doi = {10.1016/j.amc.2025.129672}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129672}, shortjournal = {Appl. Math. Comput.}, title = {Solving fuzzy non-homogeneous linear differential systems with piecewise constant arguments involving the short-memory variable-order caputo fractional derivative}, volume = {510}, year = {2026}, }
• (2026). Basket options with volatility skew: Calibrating a local volatility model by sample rearrangement. AMC, 510, 129669. (www)

  @article{ , abstract = {The pricing of derivatives tied to baskets of assets demands a sophisticated framework that aligns with the available market information to capture the intricate non-linear dependency structure among the assets. We describe the dynamics of the multivariate process of constituents with a copula model and propose an efficient method to extract the dependency structure from the market. The proposed method generates coherent sets of samples of the constituents process through systematic sampling rearrangement. These samples are then utilized to calibrate a local volatility model (LVM) of the basket process, which is used to price basket derivatives. We show that the method is capable of efficiently pricing basket options based on a large number of basket constituents, accomplishing the calibration process within a matter of seconds, and achieving near-perfect calibration to the index options of the market.}, archive = {J_AMC}, author = {Nicola F. Zaugg and Lech A. Grzelak}, doi = {10.1016/j.amc.2025.129669}, journal = {Applied Mathematics and Computation}, month = {2}, pages = {129669}, shortjournal = {Appl. Math. Comput.}, title = {Basket options with volatility skew: Calibrating a local volatility model by sample rearrangement}, volume = {510}, year = {2026}, }

ARTMED - 15

• (2025). The interpretable deep learning framework and validation for seizure detection in pediatric electroencephalography: An improved accuracy and performance analysis. ARTMED, 170, 103276. (www)

  @article{ , abstract = {This study proposes an interpretable deep learning framework and compares the two novel models. A fully convolutional network with squeeze-and-excitation modules (SE-FCN) is designed to enhance spatial sensitivity and retain temporal resolution. In addition, a transformer-based model (TransNet) is developed to capture temporal and channel-wise dependencies via self-attention. These two models output channel saliency weights to the EEG electrode space and generate heatmaps for inferring potential epileptogenic zones. Deep learning primarily adopts convolutional neural networks (CNNs) or sequence generation networks (SGNs) and faces the limitations. For instance, CNN-based models often lack hierarchical modeling and fail to quantify channel-wise contributions, hindering spatial localization. SGN-based models struggle to capture complex spatiotemporal dependencies and typically lack adaptive attention tailored to electroencephalography (EEG) characters. Epileptic seizure detection is vital for effective clinical intervention and existing methods operated as black boxes, limiting clinical interpretability. This study evaluates the models on the CHB-MIT pediatric EEG dataset using a subject-independent cross-validation protocol. SE-FCN achieves an AUC of 0.89 and accuracy of 86.7 %, while TransNet achieves an AUC of 0.92 and accuracy of 86.4 %. Saliency maps from both models demonstrate high consistency and enable categorization of 22 patients into five groups based on inferred seizure origins.}, archive = {J_ARTMED}, author = {Yu Zhou and Yuxin Gao and Qiang Li and Ruiheng Wu and Aiping Yang and Ming-Lang Tseng}, doi = {10.1016/j.artmed.2025.103276}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103276}, shortjournal = {Artif. Intell. Med.}, title = {The interpretable deep learning framework and validation for seizure detection in pediatric electroencephalography: An improved accuracy and performance analysis}, volume = {170}, year = {2025}, }
• (2025). A novel memory interaction neural network for multi-label drug–drug interaction prediction with neighbor importance sampling. ARTMED, 170, 103275. (www)

  @article{ , abstract = {Co-administration of multiple drugs can frequently cause drug–drug interactions (DDIs), including adverse drug reactions (ADRs) that may increase the likelihood of morbidity and mortality. Identifying potential DDIs presents a significant challenge, due to the complexity of pharmacology. Recent advances in knowledge graphs have contributed to DDI prediction by providing a robust framework for representing various relationships between drugs and other entities, such as proteins, diseases, and drug attributes. However, current network-based models often fail to uncover interaction information among DDI triplets, as they typically encode triplets independently. Additionally, uniform sampling methods may overlook differences in neighboring node properties. In this work, we propose a novel memory interaction neural network for DDI prediction, which integrates drug molecular sequences with semantic information from the drug knowledge graph. Specifically, we introduce a neighbor importance sampling strategy that selectively samples highly connected neighbors, improving computational efficiency and reducing noise. We also design a memory interaction module that utilizes multi-head attention mechanisms and deep neural networks to capture interactions among DDI triplets. Experimental evaluation on KEGG and OGB-biokg datasets demonstrates the superiority of our model compared to classical and state-of-the-art methods in predicting DDIs. Datasets and code for this proposed DDIs prediction model are freely accessible at https://github.com/wj1108114106/Multi-label-DDIs .}, archive = {J_ARTMED}, author = {Jing Wang and Runzhi Li and Shuo Zhang and YunLi Xing and Siyu Yan and Lihong Ma}, doi = {10.1016/j.artmed.2025.103275}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103275}, shortjournal = {Artif. Intell. Med.}, title = {A novel memory interaction neural network for multi-label drug–drug interaction prediction with neighbor importance sampling}, volume = {170}, year = {2025}, }
• (2025). Types, functions, and mechanisms of machine learning for personalizing smoking cessation interventions: A systematic scoping review. ARTMED, 170, 103274. (www)

  @article{ , abstract = {Purpose Artificial intelligence can realize personalization. This systematic scoping review provides the types, functions, and mechanisms of machine learning (ML) for personalizing smoking cessation interventions. Methodology We searched fourteen databases including PubMed, CINAHL, EMBASE, the Cochrane Library, IEEE Xplore, PsycINFO, Scopus, Web of Science, AAAI, ACM Digital Library, ArXIV, Mednar, ProQuest, and Science.gov . We selected 98 articles from 4073 records that met the criteria. Two independent reviewers screened and selected the articles. Two reviewers extracted the data using a self-developed data charting form independently. Results The findings are reported in narrative syntheses, tables, and figures. The types of ML included artificial neural networks, Bayesian algorithms, clustering algorithms, decision tree algorithms, deep learning (DL) algorithms, ensemble algorithms, linear classifiers, others, and unspecified. The most common ML technique used was supervised learning (81 %), and the ML functions included (1) message tailoring (17 %), (2) prediction and detection of smoking events (34 %), (3) social media surveillance (14 %), (4) predictive models (24 %), and (5) biomarker analysis (10 %). The ML mechanisms involved the following sequence: data input, data preprocessing, feature extraction and selection, training and validation, and data output. Conclusion This review is the first to describe the potential use of ML for personalizing smoking cessation interventions. We provide recommendations for future research by identifying the limitations and gaps in the studies. Future studies should refine, validate, and test ML models using robust experimental methods to conclude their effectiveness.}, archive = {J_ARTMED}, author = {Yu Jie Xavia Ng and Shing Hui Reina Cheong and Wen Wei Ang and Ying Lau and Siew Tiang Lau}, doi = {10.1016/j.artmed.2025.103274}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103274}, shortjournal = {Artif. Intell. Med.}, title = {Types, functions, and mechanisms of machine learning for personalizing smoking cessation interventions: A systematic scoping review}, volume = {170}, year = {2025}, }
• (2025). Decoding the cortical responses to mechanical wrist perturbations: A two-step shared structure NARX method. ARTMED, 170, 103273. (www)

  @article{ , abstract = {Shared structure nonlinear autoregressive with exogenous input (NARX) model is a promising tool for exploring cortical responses mechanism to external stimuli, essential for advancing our understanding of brain function and developing methods for direct brain information encoding. In this paper, we proposed a two-step method to overcome limitations in existing method, which neglect data relationships and rely on a greedy search for regression terms, leading to less accurate models. In our approach, data from multiple trials are concatenated, and then the orthogonal forward regression (OFR) algorithm identifies model terms in first step, enhancing inter-trial connections and establishing a preliminary model for each subject. Shared model terms across subjects are then used to construct a general target model. Next, non-shared regression terms that best represent population-level information are identified, using adaptive multi-population genetic algorithms, and use to enhance the target models' descriptive power. Simulations results show significant competitiveness in terms of accuracy as compared to other state-of-the-art methods. When applied to real electroencephalography signals under mechanical disturbance, structural and parameter analysis revealed consistent neural response patterns across subjects, with subject-specific responses likely stemming from muscle feedback. Frequency response analysis further suggests that the brain may generate motor inhibition signals based on sensory inputs to maintain a pre-disturbance resting state. These findings provide valuable insights into cortical response mechanisms and have potential implications for future brain information encoding research.}, archive = {J_ARTMED}, author = {Nan Zheng and Yurong Li and Wuxiang Shi and Jiyu Tan}, doi = {10.1016/j.artmed.2025.103273}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103273}, shortjournal = {Artif. Intell. Med.}, title = {Decoding the cortical responses to mechanical wrist perturbations: A two-step shared structure NARX method}, volume = {170}, year = {2025}, }
• (2025). AI-driven dynamic grouping for adaptive clinical trials: Rethinking randomization in precision medicine. ARTMED, 170, 103272. (www)

  @article{ , abstract = {Integrating artificial intelligence into biomedical human subjects research is transforming traditional experimental paradigms. This perspective introduces the concept of “dynamic grouping,” wherein artificial intelligence (AI) systems continuously reassign participants across experimental conditions based on real-time biomarker data and clinical response patterns. Unlike traditional biomedical research designs that rely on fixed treatment and control groups, dynamic grouping allows participant assignments to evolve throughout the study. We examine the ethical implications, methodological challenges, and research opportunities associated with this paradigm, particularly in clinical trials, precision medicine, and digital therapeutics. To support this analysis, we present three computational simulations that quantify its impact: (i) a heterogeneity simulation demonstrating how patient variability affects the advantage of dynamic grouping, (ii) a statistical power analysis showing potential sample size reductions in adaptive designs, and (iii) a clinical outcome distribution analysis highlighting how dynamic grouping reduces negative treatment outcomes and optimizes patient responses. Our findings suggest that dynamic grouping can improve treatment effectiveness, enhance resource allocation, and increase statistical efficiency, although it also raises new challenges for causal inference, informed consent, and regulatory oversight. As AI continues to reshape medical research, adapting ethical and methodological frameworks will be essential for its responsible implementation.}, archive = {J_ARTMED}, author = {Madhur Mangalam}, doi = {10.1016/j.artmed.2025.103272}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103272}, shortjournal = {Artif. Intell. Med.}, title = {AI-driven dynamic grouping for adaptive clinical trials: Rethinking randomization in precision medicine}, volume = {170}, year = {2025}, }
• (2025). H-SynEx: Using synthetic images and ultra-high resolution ex vivo MRI for hypothalamus subregion segmentation. ARTMED, 170, 103271. (www)

  @article{ , abstract = {The hypothalamus is a small structure located in the center of the brain and is involved in significant functions such as sleeping, temperature, and appetite control. Various neurological disorders are also associated with hypothalamic abnormalities. Automated image analysis of this structure from brain MRI is thus highly desirable to study the hypothalamus in vivo . However, most of the automated segmentation tools currently available focus exclusively on T1w images. In this study, we introduce H-SynEx, a machine learning method for automated segmentation of hypothalamic subregions that generalizes across different MRI sequences and resolutions without retraining. H-synEx was trained with synthetic images built from label maps derived from ultra-high resolution ex vivo MRI scans, allowing finer-grained manual segmentation when compared with 1 mm isometric in vivo images. We validated our method using Dice Coefficient (DSC) and Average Hausdorff distance (AVD) across in vivo images from six different datasets with six different MRI sequences (T1, T2, proton density, quantitative T1, fractional anisotropy, and FLAIR). Statistical analysis compared hypothalamic subregion volumes in controls, Alzheimer’s disease (AD), and behavioral variant frontotemporal dementia (bvFTD) subjects using the Area Under the Receiver Operating Characteristic curve (AUROC) and the Wilcoxon rank sum test. Our results show that H-SynEx successfully leverages information from ultra-high resolution scans to segment in vivo from different MRI sequences. Our automated segmentation was able to discriminate controls versus patients with Alzheimer’s disease on FLAIR images with 5 mm spacing. H-SynEx is openly available at https://github.com/liviamarodrigues/hsynex .}, archive = {J_ARTMED}, author = {Livia Rodrigues and Martina Bocchetta and Oula Puonti and Douglas Greve and Ana Carolina Londe and Marcondes França and Simone Appenzeller and Juan Eugenio Iglesias and Leticia Rittner}, doi = {10.1016/j.artmed.2025.103271}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103271}, shortjournal = {Artif. Intell. Med.}, title = {H-SynEx: Using synthetic images and ultra-high resolution ex vivo MRI for hypothalamus subregion segmentation}, volume = {170}, year = {2025}, }
• (2025). Mining multi-electrode and multi-wave electroencephalogram based time-interval temporal patterns for improved classification capabilities and explainability. ARTMED, 170, 103269. (www)

  @article{ , abstract = {Brain-computer interface (BCI) systems, and particularly electroencephalogram (EEG) based BCI systems, have become more widely used in recent years and are utilized in various applications and domains ranging from medicine and marketing to games and entertainment. While different algorithms have been used to analyze EEG data and enable its classification, existing algorithms have two main drawbacks; both their classification and explainability capabilities are limited. Lacking in explainability, they cannot indicate which electrodes and waves led to a classification decision or explain how areas and frequencies of the brain's activity correlate to a specific task. In this study, we propose a novel extension for the time-interval temporal patterns mining algorithms aimed at enhancing the data mining process by enabling a richer set of patterns to be learned from the EEG data, thereby contributing to improved classification and explainability capabilities. The extended algorithm is designed to capture and leverage the unique nature of EEG data by decomposing it into different brain waves and modeling the relations among them and between different electrodes. Our evaluation of the proposed extended algorithm on multiple learning tasks and three EEG datasets demonstrated the extended algorithm's ability to mine richer patterns that improve the classification performance by 4–11 % based on the Area-Under the receiver operating characteristic Curve (AUC) metric, compared to the original version of the algorithm. Moreover, the algorithm was shown to shed light on the areas and frequencies of the brain's activity that are correlated with specific tasks.}, archive = {J_ARTMED}, author = {Ofir Landau and Nir Nissim}, doi = {10.1016/j.artmed.2025.103269}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103269}, shortjournal = {Artif. Intell. Med.}, title = {Mining multi-electrode and multi-wave electroencephalogram based time-interval temporal patterns for improved classification capabilities and explainability}, volume = {170}, year = {2025}, }
• (2025). A survey for large language models in biomedicine. ARTMED, 170, 103268. (www)

  @article{ , abstract = {Recent breakthroughs in large language models (LLMs) offer unprecedented natural language understanding and generation capabilities. However, existing surveys on LLMs in biomedicine often focus on specific applications or model architectures, lacking a comprehensive analysis that integrates the latest advancements across various biomedical domains. This review, based on an analysis of 484 publications sourced from databases including PubMed, Web of Science, and arXiv, provides an in-depth examination of the current landscape, applications, challenges, and prospects of LLMs in biomedicine, distinguishing itself by focusing on the practical implications of these models in real-world biomedical contexts. Firstly, we explore the capabilities of LLMs in zero-shot learning across a broad spectrum of biomedical tasks, including diagnostic assistance, drug discovery, and personalized medicine, among others, with insights drawn from 137 key studies. Then, we discuss adaptation strategies of LLMs, including fine-tuning methods for both uni-modal and multi-modal LLMs to enhance their performance in specialized biomedical contexts where zero-shot fails to achieve, such as medical question answering and efficient processing of biomedical literature. Finally, we discuss the challenges that LLMs face in the biomedicine domain including data privacy concerns, limited model interpretability, issues with dataset quality, and ethics due to the sensitive nature of biomedical data, the need for highly reliable model outputs, and the ethical implications of deploying AI in healthcare. To address these challenges, we also identify future research directions of LLM in biomedicine including federated learning methods to preserve data privacy and integrating explainable AI methodologies to enhance the transparency of LLMs. As this field of LLM rapidly evolves, continued research and development are essential to fully harness the capabilities of LLMs in biomedicine while ensuring their responsible and effective deployment.}, archive = {J_ARTMED}, author = {Chong Wang and Mengyao Li and Junjun He and Zhongruo Wang and Erfan Darzi and Zan Chen and Jin Ye and Tianbin Li and Yanzhou Su and Jing Ke and Kaili Qu and Shuxin Li and Yi Yu and Pietro Liò and Tianyun Wang and Yu Guang Wang and Yiqing Shen}, doi = {10.1016/j.artmed.2025.103268}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103268}, shortjournal = {Artif. Intell. Med.}, title = {A survey for large language models in biomedicine}, volume = {170}, year = {2025}, }
• (2025). High-quality triage and diagnosis of gynecological diseases via artificial intelligence. ARTMED, 170, 103267. (www)

  @article{ , abstract = {Timely detection and diagnosis of diseases are key elements of an efficient healthcare system. In recent years, artificial intelligence (AI) has played an increasingly important role in improving the accuracy and efficiency of disease diagnosis in clinical practice. However, most existing AI systems for disease diagnosis have focused on either classifying patients into broad disease categories or diagnosing a specific disease, leaving a gap in the development of a coherent AI system for both triage and diagnosis in a department of a general hospital. In this study, we fill this gap with SmartGyne, an advanced AI system that can achieve high-quality triage and diagnosis for a full spectrum of gynecological diseases. By extracting useful clinical evidence for diagnosis from a large amount of electronic medical records, SmartGyne establishes an effective framework to integrate real-world clinical evidence and knowledge into a coherent AI system that can effectively handle a full spectrum of complex diseases in a department of a general hospital. Validation experiments demonstrated that SmartGyne achieved an overall accuracy of 80.1 % in triage for gynecological diseases, and 99.4 % in diagnosis for a gynecological subspecialty. In comparison with human physicians, SmartGyne showed competitive triage and diagnostic performance, and improved consultation efficiency and accuracy for physicians with limited specialized experience. These results show that SmartGyne achieves high-quality triage and diagnosis, holding the potential to improve the efficiency of the healthcare system in China, as well as other countries lacking professional gynecologists.}, archive = {J_ARTMED}, author = {Linru Fu and Che Wang and Zhaoyang Liu and Changzai Pan and Zhe Du and Zhijing Sun and Lan Zhu and Ke Deng}, doi = {10.1016/j.artmed.2025.103267}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103267}, shortjournal = {Artif. Intell. Med.}, title = {High-quality triage and diagnosis of gynecological diseases via artificial intelligence}, volume = {170}, year = {2025}, }
• (2025). Improving prototypical parts abstraction for case-based reasoning explanations designed for the kidney stone type recognition. ARTMED, 170, 103266. (www)

  @article{ , abstract = {The in-vivo identification of the kidney stone types during an ureteroscopy would be a major medical advance in urology, as it could reduce the time of the tedious renal calculi extraction process, while diminishing infection risks. Furthermore, such an automated procedure would make possible to prescribe anti-recurrence treatments immediately. Nowadays, only few experienced urologists are able to recognize the kidney stone types in the images of the videos displayed on a screen during the endoscopy. This visual recognition by urologists is also highly operator dependent. Thus, several deep learning (DL) models have recently been proposed to automatically recognize the kidney stone types using ureteroscopic images. However, these DL models are of black box nature and do not establish the relationship of the visual features they used to take the decision with the color, texture and morphological features visually analyzed in biological laboratories to determine the type of extracted kidney stone fragments using the reference morphoconstitutional analysis (MCA) procedure. This contribution proposes a case-based reasoning DLmodel which uses prototypical parts (PPs) and generates local and global descriptors. The PPs encode for each class (i.e., kidney stone type) visual feature information (hue, saturation, intensity and textures) similar to that used by biologists during MCA. The PPs are optimally generated due a new loss function used during the model training. Moreover, the local and global descriptors of PPs allow to explain the decisions (“what” information, “where in the images”) in an understandable way for biologists and urologists. The proposed DL model has been tested on a database including images of the six most widespread kidney stone types in industrialized countries. The overall average classification accuracy was 90 . 37 ± 0 . 6 % . When comparing this results with that of the eight other DL models of the kidney stone state-of-the-art, it can be seen that the valuable gain in explanability was not reached at the expense of accuracy which was even slightly increased with respect to that ( 88 . 2 ± 2 . 1 % ) of the best method of the literature. These promising and interpretable results also encourage urologists to put their trust in AI-based solutions.}, archive = {J_ARTMED}, author = {Daniel Flores-Araiza and Francisco Lopez-Tiro and Clément Larose and Salvador Hinojosa and Andres Mendez-Vazquez and Miguel Gonzalez-Mendoza and Gilberto Ochoa-Ruiz and Christian Daul}, doi = {10.1016/j.artmed.2025.103266}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103266}, shortjournal = {Artif. Intell. Med.}, title = {Improving prototypical parts abstraction for case-based reasoning explanations designed for the kidney stone type recognition}, volume = {170}, year = {2025}, }
• (2025). Medical multimodal foundation models in clinical diagnosis and treatment: Applications, challenges, and future directions. ARTMED, 170, 103265. (www)

  @article{ , abstract = {Recent advancements in deep learning have significantly revolutionized the field of clinical diagnosis and treatment, offering novel approaches to improve diagnostic precision and treatment efficacy across diverse clinical domains, thus driving the pursuit of precision medicine. The growing availability of multi-organ and multimodal datasets has accelerated the development of large-scale Medical Multimodal Foundation Models (MMFMs). These models, known for their strong generalization capabilities and rich representational power, are increasingly being adapted to address a wide range of clinical tasks, from early diagnosis to personalized treatment strategies. This review offers a comprehensive analysis of recent developments in MMFMs, focusing on three key aspects: datasets, model architectures, and clinical applications. We also explore the challenges and opportunities in optimizing multimodal representations and discuss how these advancements are shaping the future of healthcare by enabling improved patient outcomes and more efficient clinical workflows.}, archive = {J_ARTMED}, author = {Kai Sun and Siyan Xue and Fuchun Sun and Haoran Sun and Yu Luo and Ling Wang and Siyuan Wang and Na Guo and Lei Liu and Tian Zhao and Xinzhou Wang and Lei Yang and Shuo Jin and Jun Yan and Jiahong Dong}, doi = {10.1016/j.artmed.2025.103265}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103265}, shortjournal = {Artif. Intell. Med.}, title = {Medical multimodal foundation models in clinical diagnosis and treatment: Applications, challenges, and future directions}, volume = {170}, year = {2025}, }
• (2025). An end-to-end solution for out-of-hospital emergency medical dispatch triage based on multimodal and continual deep learning. ARTMED, 170, 103264. (www)

  @article{ , abstract = {The objective of this study was to build a multimodal, multitask predictive model—named E2eDeepEMC 2 —to improve out-of-hospital emergency incident severity assessments while coping with shifts in data distributions over time. We drew on 2 054 694 independent incidents recorded by the Valencian emergency medical dispatch service between 2009 and 2019 (excluding 2013), combining demographic, temporal, clinical and free-text inputs. To handle temporal drift, our model integrates continual learning strategies and comprises three encoder modules (for context, clinical data and text), whose outputs are merged to predict the life-threatening level, admissible response delay and emergency system jurisdiction. Compared with the Valencian Region’s existing in-house triage protocol, E2eDeepEMC 2 achieved absolute F1-score gains of 18.46% for life-threatening level, 25.96% for response delay and 3.63% for jurisdiction. Compared to non-continual learning baselines, it also outperformed them by 3.04%, 9.66% and 0.58%, respectively. Deployment of E2eDeepEMC 2 is currently underway in the Valencian Region, underscoring its practical impact on real-world emergency dispatch decision-making.}, archive = {J_ARTMED}, author = {Pablo Ferri and Carlos Sáez and Antonio Félix-De Castro and Purificación Sánchez-Cuesta and Juan M. García-Gómez}, doi = {10.1016/j.artmed.2025.103264}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103264}, shortjournal = {Artif. Intell. Med.}, title = {An end-to-end solution for out-of-hospital emergency medical dispatch triage based on multimodal and continual deep learning}, volume = {170}, year = {2025}, }
• (2025). LoRA-PT: Low-rank adapting UNETR for hippocampus segmentation using principal tensor singular values and vectors. ARTMED, 170, 103254. (www)

  @article{ , abstract = {The hippocampus is an important brain structure involved in various psychiatric disorders, and its automatic and accurate segmentation is vital for studying these diseases. Recently, deep learning-based methods have made significant progress in hippocampus segmentation. However, training deep neural network models requires substantial computational resources, time, and a large amount of labeled training data, which is frequently scarce in medical image segmentation. To address these issues, we propose LoRA-PT, a novel parameter-efficient fine-tuning (PEFT) method that transfers the pre-trained UNETR model from the BraTS2021 dataset to the hippocampus segmentation task. Specifically, LoRA-PT divides the parameter matrix of the transformer structure into three distinct sizes, yielding three third-order tensors. These tensors are decomposed using tensor singular value decomposition to generate low-rank tensors consisting of the principal singular values and vectors, with the remaining singular values and vectors forming the residual tensor. During fine-tuning, only the low-rank tensors (i.e., the principal tensor singular values and vectors) are updated, while the residual tensors remain unchanged. We validated the proposed method on three public hippocampus datasets, and the experimental results show that LoRA-PT outperformed state-of-the-art PEFT methods in segmentation accuracy while significantly reducing the number of parameter updates. Our source code is available at https://github.com/WangangCheng/LoRA-PT/tree/LoRA-PT .}, archive = {J_ARTMED}, author = {Guanghua He and Wangang Cheng and Hancan Zhu and Gaohang Yu}, doi = {10.1016/j.artmed.2025.103254}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103254}, shortjournal = {Artif. Intell. Med.}, title = {LoRA-PT: Low-rank adapting UNETR for hippocampus segmentation using principal tensor singular values and vectors}, volume = {170}, year = {2025}, }
• (2025). MMSupcon: An image fusion-based multi-modal supervised contrastive method for brain tumor diagnosis. ARTMED, 170, 103253. (www)

  @article{ , abstract = {The diagnosis of brain tumors is pivotal for effective treatment, with MRI serving as a commonly used non-invasive diagnostic modality in clinical practices. Fundamentally, brain tumor diagnosis is a type of pattern recognition task that requires the integration of information from multi-modal MRI images. However, existing fusion strategies are hindered by the scarcity of multi-modal imaging samples. In this paper, we propose a new training paradigm tailored for the scenario of multi-modal imaging in brain tumor diagnosis, called multi-modal supervised contrastive learning method (MMSupcon). This method significantly enhances diagnostic accuracy through two key components: multi-modal medical image fusion and multi-modal supervised contrastive loss. First, the fusion component integrates complementary imaging modalities to generate information-rich samples. Second, by introducing fused samples to guide original samples in learning feature consistency or inconsistency among classes, our loss component effectively preserves the integrity of cross-modal information while maintaining the distinctiveness of individual modalities. Finally, MMSupcon is validated on a real-world brain tumor dataset collected from Beijing Tiantan Hospital, achieving state-of-the-art performance. Furthermore, additional experiments on two public BraTS glioma classification datasets also demonstrate our substantial performance improvements. The source code is released at https://github.com/hywang02/MMSupcon .}, archive = {J_ARTMED}, author = {Haoyu Wang and Jing Zhang and Siying Wu and Haoran Wei and Xun Chen and Yunwei Ou and Xiaoyan Sun}, doi = {10.1016/j.artmed.2025.103253}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103253}, shortjournal = {Artif. Intell. Med.}, title = {MMSupcon: An image fusion-based multi-modal supervised contrastive method for brain tumor diagnosis}, volume = {170}, year = {2025}, }
• (2025). Pathway information on methylation analysis using deep neural network (PROMINENT): An interpretable deep learning method with pathway prior for phenotype prediction using gene-level DNA methylation. ARTMED, 170, 103236. (www)

  @article{ , abstract = {Background DNA methylation is a key epigenetic marker that influences gene expression and phenotype regulation, and is affected by both genetic and environmental factors. Traditional linear regression methods such as elastic nets have been employed to assess the cumulative effects of multiple DNA methylation markers on phenotypes. However, these methods often fail to capture the complex nonlinear nature of the data. Recent deep learning approaches, such as MethylNet, have improved the prediction accuracy but lack interpretability and efficiency. Findings To address these limitations, we introduced P athway Info r mati o n on M ethylat i on Analysis using a Deep Ne ural N e t work (PROMINENT), a novel interpretable deep learning method that integrates gene-level DNA methylation data with biological pathway information for phenotype prediction. PROMINENT enhances interpretability and prediction accuracy by incorporating gene- and pathway-level priors from databases such as Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG). It employs SHapley Additive exPlanations (SHAP) to prioritize significant genes and pathways. Evaluated across various datasets, childhood asthma, idiopathic pulmonary fibrosis (IPF), and first-episode psychosis (FEP)—PROMINENT consistently outperformed existing methods in terms of prediction accuracy and computational efficiency. PROMINENT also identified crucial genes and pathways involved in disease mechanisms. Conclusions PROMINENT represents a significant advancement in leveraging DNA methylation data for phenotype prediction, offering both high accuracy and interpretability within reasonable computational time. This method holds promise for elucidating the epigenetic underpinnings of complex diseases and enhancing the utility of DNA methylation data in biomedical research.}, archive = {J_ARTMED}, author = {Soyeon Kim and Laizhi Zhang and Yidi Qin and Rebecca I. Caldino Bohn and Hyun Jung Park}, doi = {10.1016/j.artmed.2025.103236}, journal = {Artificial Intelligence in Medicine}, month = {12}, pages = {103236}, shortjournal = {Artif. Intell. Med.}, title = {Pathway information on methylation analysis using deep neural network (PROMINENT): An interpretable deep learning method with pathway prior for phenotype prediction using gene-level DNA methylation}, volume = {170}, year = {2025}, }

ASOC - 228

• (2025). Detail-aware semantic segmentation network for brain tumor MRI images combining multi-frequency directional filtering and lifting wavelets. ASOC, 185, 113969. (www)

  @article{ , abstract = {Brain tumors segmentation in Magnetic Resonance Imaging (MRI) images poses significant challenges owing to the uncertain location and size of the tumors, the difficulty in describing their boundaries, and the fuzzy demarcation of diseased tissues. Although U-Net and its recent variants have emerged as leading models for semantic segmentation in medical imaging, they still face structural limitations. These limitations cause the erosion of detail information during downsampling and poor performance in segmenting small lesions when handling targets of varying sizes, indicating a lack of detail handling capability. To counteract these issues, we designed a segmentation model that enhances detail features using frequency information. To reduce the loss of feature information during downsampling, we developed a downsampling module based on lifting wavelets. By lifting wavelets to group and integrate features according to frequency from high to low, we reduce feature resolution while enhancing information transmission and minimizing feature information loss. In our designed multi-frequency directional filtering edge feature extraction module, we extract low-frequency and high-frequency features and construct a dual-channel multi-directional filtering combination. This combination extracts directional information from low-frequency and high-frequency features separately, increasing the multi-angle directional information of the features and enriching the detailed information such as direction and position within the features. On the BraTS2018, BraTS2020, and BraTS2024 brain tumor datasets, our model demonstrated optimal results compared to 14 other advanced models. The average Dice Similarity Coefficients are 78.48 %, 79.80 %, and 74.35 %, while the 95th percentile Hausdorff Distances are 5.75, 6.60, and 7.72. Our code link is https://github.com/Eric-H8/BraTS_Seg_Model .}, archive = {J_ASOC}, author = {Xin Hua and Zhijiang Du and Hongjian Yu and Zibo Li and Qiaohui Lu and Hui Zhao}, doi = {10.1016/j.asoc.2025.113969}, journal = {Applied Soft Computing}, month = {12}, pages = {113969}, shortjournal = {Appl. Soft. Comput.}, title = {Detail-aware semantic segmentation network for brain tumor MRI images combining multi-frequency directional filtering and lifting wavelets}, volume = {185}, year = {2025}, }
• (2025). Extracting knowledge from limited data: An updated review of data-driven and model-driven few-shot learning for agriculture. ASOC, 185, 113968. (www)

  @article{ , abstract = {Deep learning has demonstrated considerable success in agricultural applications. However, its conventional implementations heavily depend on large-scale labelled datasets—a requirement that is often impractical in agriculture due to data scarcity, high annotation costs, or environmental variability. While insufficient training data can significantly limit the performance of standard deep learning models, Few-Shot Learning (FSL) has emerged as a transformative paradigm, enabling robust model training with minimal labelled samples by utilising limited data for training instead. Despite its potential, a critical review assessing how FSL addresses expert system challenges in agriculture remains notably absent. This paper attempts to fill this void by presenting an updated comprehensive review of FSL's applications in agriculture. We categorise FSL methodologies into two primary approaches: data processing-driven and model learning-driven. Data processing–driven approaches address data scarcity by enriching representational diversity through synthetic samples generated with models such as generative adversarial networks, or by transferring knowledge from related domains to improve generalisation. In contrast, model learning–driven strategies confront the same challenge through specialised architectures and optimisation techniques that enable effective generalisation from limited samples. Within this taxonomy, data processing–driven paradigms include transfer learning and generative artificial intelligence, while model learning–driven paradigms cover metric learning methods such as Siamese or prototypical networks, together with model-based and optimisation approaches designed for efficient generalisation. Our analysis pinpoints cutting-edge technologies within each sector, shedding light on overlooked areas and opportunities where FSL can harness limited data to yield promising outcomes when used to solve problems in agriculture.}, archive = {J_ASOC}, author = {Kam Meng Goh and Usman Ullah Sheikh and Jun Kit Chaw and Weng Kin Lai and Weng Chun Tan and Santhi Krishnamoorthy}, doi = {10.1016/j.asoc.2025.113968}, journal = {Applied Soft Computing}, month = {12}, pages = {113968}, shortjournal = {Appl. Soft. Comput.}, title = {Extracting knowledge from limited data: An updated review of data-driven and model-driven few-shot learning for agriculture}, volume = {185}, year = {2025}, }
• (2025). Coverage exploration of unknown obstacle-cluttered environments using a swarm of ground robots. ASOC, 185, 113964. (www)

  @article{ , abstract = {This paper introduces a coverage exploration algorithm for unknown obstacle-cluttered environments using a swarm of ground robots. A key contribution of this work is the proposed fitness function, which balances multiple exploration objectives and encourages robots to disperse effectively, avoiding excessive overlapping visits. The robots are assumed to start from a single corner of the environment, reflecting practical situations where pre-distributing them is not feasible. This setup highlights a key feature of the algorithm, as it enables self-organization and effective distribution of the robots throughout the environment. The robustness of the method is demonstrated through experiments in various environmental setups, showing its resilience to different obstacle structures and reliable performance across diverse scenarios. The approach also leverages the benefits of swarm behavior, where an increasing number of robots improves exploration efficiency through enhanced collaboration and coverage. The algorithm is evaluated against a swarm random walk approach and two multi-robot meta-heuristic methods, significantly outperforming them in terms of coverage efficiency and robustness.}, archive = {J_ASOC}, author = {Khalil Al-rahman Youssefi and Wilfried Elmenreich}, doi = {10.1016/j.asoc.2025.113964}, journal = {Applied Soft Computing}, month = {12}, pages = {113964}, shortjournal = {Appl. Soft. Comput.}, title = {Coverage exploration of unknown obstacle-cluttered environments using a swarm of ground robots}, volume = {185}, year = {2025}, }
• (2025). WamGLM: A multimodal large-scale language model for wafer map defect information in-depth query through multi-turn dialogue based on prototypical supervised contrastive learning. ASOC, 185, 113962. (www)

  @article{ , abstract = {To ensure production efficiency and process stability in semiconductor manufacturing, it is of critical importance to detect wafer map defects and perform information query for tracing and solving problems during the manufacturing process. Numerous vision models based on deep learning have been successfully applied to wafer map defect recognition (WMDR), yielding remarkable results. However, the dynamic and in-depth querying of wafer map defect information remains relatively underexplored. Leveraging the rapid advancements in multimodal large language models (MLLMs), this paper proposes a novel approach for wafer map defect information query (WMDIQ). First, following the paradigm of employing cross-modal alignment model to bridge vision and language models, an end-to-end response MLLM: general language model for wafer map (WamGLM), is constructed for WMDIQ. Concurrently, by designing an interactive dialogue framework between large language models (LLMs), the first large-scale multi-turn dialogue dataset: visual multi-turn question answering dataset for wafer map defects (WaferMapVMQA Dataset), is constructed for wafer map defect analysis. Subsequently, WamGLM is trained using a two-stage fine-tuning strategy. In the first stage, a visual fine-tuning method based on prototypical supervised contrastive learning (PSCL) is introduced to enhance the intra-class compactness and inter-class separability of defect features. In the second stage, language fine-tuning is conducted using the WaferMapVMQA Dataset to infuse specialized knowledge into WamGLM. To validate the effectiveness and superiority of the proposed method, experiments are conducted on a real wafer map dataset. The results demonstrate that the proposed method significantly outperforms other methods in both defect recognition performance and information query response performance. Our code is available at: https://github.com/ZihaoLei/WamGLM .}, archive = {J_ASOC}, author = {Shulong Gu and Zihao Lei and Guangrui Wen and Quanning Xu and Zhaojun Steven Li and Xuefeng Chen and Chunsheng Yang}, doi = {10.1016/j.asoc.2025.113962}, journal = {Applied Soft Computing}, month = {12}, pages = {113962}, shortjournal = {Appl. Soft. Comput.}, title = {WamGLM: A multimodal large-scale language model for wafer map defect information in-depth query through multi-turn dialogue based on prototypical supervised contrastive learning}, volume = {185}, year = {2025}, }
• (2025). Multi-criteria linguistic optimization for covert communication in secure LLM-based steganography. ASOC, 185, 113960. (www)

  @article{ , abstract = {This paper presents a novel framework for covert communication through secure steganography using large language models (LLMs). Our approach leverages multi-criteria linguistic optimization to encode secret information directly into stylistic features of auto-regressively generated text. This strategy balances embedding capacity with naturalness and coherence. The secret message is partitioned into fixed-size blocks. Each block is embedded into binary stylistic feature vectors via a surjective linear mapping, which introduces redundancy. This redundancy enables the use of a history-aware cost function that selects stylistic vectors to minimize abrupt transitions and preserve fluency across sentences. Candidate sentences are generated by prompting LLMs with contextual and stylistic constraints. Rejection sampling then ensures exact feature matching and high linguistic quality. Experimental evaluation in multiple LLMs, diverse text contexts, and parameter settings demonstrates effective embedding capacities of up to 0.30 bits per token while maintaining strong linguistic naturalness, validated through perplexity, lexical diversity, readability, and a linguistic acceptability metric. Importantly, decoding recovers the full secret with zero error under ideal conditions. This confirms the reliability of the method. The current work focuses on embedding efficiency and imperceptibility. Robustness against active text alterations and formal undetectability assessments remain open challenges for future research. The proposed multi-criteria linguistic optimization framework offers a promising avenue for advanced covert communication by harmonizing secure information embedding with fluent, human-like language generation.}, archive = {J_ASOC}, author = {Kamil Woźniak and Marek R. Ogiela and Lidia Ogiela}, doi = {10.1016/j.asoc.2025.113960}, journal = {Applied Soft Computing}, month = {12}, pages = {113960}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-criteria linguistic optimization for covert communication in secure LLM-based steganography}, volume = {185}, year = {2025}, }
• (2025). A granularity time series forecasting model combining three-way decision and trend information granule. ASOC, 185, 113957. (www)

  @article{ , abstract = {Long-term forecasting of time series plays a vital role across diverse applications but is challenged by error accumulation arising from recursive predictions and the insufficient retention of trend information in conventional methods. To tackle these issues, we propose a novel forecasting model based on granular time series (GTS). The model utilizes an improved L 1 -trend filtering technique to achieve optimal segmentation of information granules, preserving essential trend features. Subsequently, we introduce dual evaluation functions based on distance similarity to jointly drive the three-way decision (TWD) process for aggregating information granules, thereby effectively reducing error propagation. Finally, the aggregated granules serve as inputs to a long short-term memory (LSTM) neural network to generate accurate forecasts. In addition, the proposed model is evaluated on several real-world datasets through sensitivity and comparative analyses. The results demonstrate that the model exhibits strong performance in long-term forecasting tasks.}, archive = {J_ASOC}, author = {Jianuan Qiu and Shuhua Su and Jingjing Qian}, doi = {10.1016/j.asoc.2025.113957}, journal = {Applied Soft Computing}, month = {12}, pages = {113957}, shortjournal = {Appl. Soft. Comput.}, title = {A granularity time series forecasting model combining three-way decision and trend information granule}, volume = {185}, year = {2025}, }
• (2025). Multi-objective genetic programming for binary classification with adaptive thresholds and a generalization-optimizing fitness function. ASOC, 185, 113956. (www)

  @article{ , abstract = {Genetic programming (GP) has been widely applied to classifier construction due to its flexible representation and powerful feature construction capabilities. Existing studies have proposed various fitness functions to improve GP-based classifiers, but most of them rely on a fixed decision threshold. However, when dealing with imbalanced classification problems, a fixed threshold often biases the model toward the majority class, thereby compromising overall performance. To address this issue, in this paper, we propose a novel multi-objective GP framework for constructing binary classifiers with adaptive threshold adjustment. During evolution, the method employs Youden’s Index to dynamically adjust the threshold of each individual, enabling the classifiers to better fit the underlying data distribution. In addition, we introduce a new class separation metric, dist t , to quantify the clarity of class boundaries and enhance the generalization ability of the evolved models. The framework jointly optimizes three objectives: minority class accuracy, majority class accuracy, and the proposed dist t metric. Experiments on 14 imbalanced datasets demonstrate that our method significantly outperforms conventional single-objective GP with fixed thresholds. Further results also confirm the positive impact of the proposed dist t metric on classification performance. Compared to seven existing GP algorithms and five traditional machine learning classifiers, our approach achieves superior overall performance and better generalization ability.}, archive = {J_ASOC}, author = {Minghui Bai and Yuan Gao and Xiaoying Gao and Jianbin Ma}, doi = {10.1016/j.asoc.2025.113956}, journal = {Applied Soft Computing}, month = {12}, pages = {113956}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-objective genetic programming for binary classification with adaptive thresholds and a generalization-optimizing fitness function}, volume = {185}, year = {2025}, }
• (2025). Adaptive fuzzy entropy optimization with opposition-based archimedes search for robust multilevel image segmentation. ASOC, 185, 113943. (www)

  @article{ , abstract = {Image segmentation plays a critical role in diverse computer vision applications. Multilevel thresholding (MLT) remains one of its most widely used unsupervised techniques due to its simplicity and interpretability. However, existing MLT methods often suffer from two major limitations: (1) the inability to adapt to local intensity variations and (2) the computational burden associated with high-dimensional threshold search. To address these challenges, this study proposes a novel segmentation framework that integrates a Proximity-Adaptive Fuzzy Entropy (PAFE) model with an Opposition-Based Learning-enhanced Archimedes Optimization Algorithm (OBL-EAOA). The PAFE model utilizes dynamically adjusted trapezoidal membership functions based on intensity proximity to candidate thresholds, allowing for a more adaptive and smooth entropy surface. Meanwhile, the OBL-EAOA enhances optimization performance through opposition-based learning and adaptive parameter control, improving exploration diversity and convergence speed. The proposed PAFE-EAOA framework is validated on two benchmark datasets, BSD500 and PASCAL VOC 2012, using five standard metrics: PSNR, SSIM, FSIM, SNR, and computation time. Compared with several state-of-the-art methods including Kapur Entropy (KE)-EAOA, Fuzzy Entropy (FE)-EAOA, Patch-Levy-Based Bees Algorithm (PLBA), Marine Predators Algorithm (MPA), Improved Grey Wolf Optimizer (IGWO), and standard Archimedes Optimization Algorithm (AOA), the proposed approach consistently achieves superior segmentation quality. Notably, it reduces computation time by up to 60 % and achieves statistically significant improvements, as confirmed by the Wilcoxon signed-rank test. These results demonstrate the framework’s robustness, scalability, and effectiveness for real-world MLT-based image segmentation.}, archive = {J_ASOC}, author = {Anusha Ganesan and Sungho Kim and Ganesan Nagabushnam}, doi = {10.1016/j.asoc.2025.113943}, journal = {Applied Soft Computing}, month = {12}, pages = {113943}, shortjournal = {Appl. Soft. Comput.}, title = {Adaptive fuzzy entropy optimization with opposition-based archimedes search for robust multilevel image segmentation}, volume = {185}, year = {2025}, }
• (2025). DT and LLM driven intelligent maintenance system for L-DED and DAG-based LLM fault diagnosis evaluation framework. ASOC, 185, 113942. (www)

  @article{ , abstract = {Metal additive manufacturing (AM) has revolutionized industries such as aerospace and automotive manufacturing due to its ability to rapidly prototype complex structures. Laser Directed Energy Deposition (L-DED) is a key AM technique, offering high deposition rates and superior mechanical properties. However, the inherent complexity and high cost of L-DED equipment demand reliable maintenance management to minimize downtime. Traditional maintenance approaches struggle to keep pace with escalating production demands and to cope with growing equipment complexity. To address this, we propose a dual-driven intelligent maintenance system for L-DED, integrating Digital Twins (DT) and Large Language Models (LLMs). The system features a comprehensive DT framework that synchronizes the virtual entity with the physical one in real time, it also incorporates an intelligent maintenance Q&A assistant powered by Retrieval-Augmented Generation (RAG), leveraging L-DED maintenance knowledge bases to provide accurate operational support. Additionally, we propose a Directed Acyclic Graphs (DAG)-based framework to assess LLMs’ ability to guide users through complete fault diagnosis. Our work aims to enhance the reliability and efficiency of L-DED maintenance through advanced digital technologies, ultimately improving productivity and reducing downtime in additive manufacturing.}, archive = {J_ASOC}, author = {Jian Tang and Shitong Peng and Jianan Guo and Danya Song and Dongna Gao and Weiwei Liu and Fengtao Wang}, doi = {10.1016/j.asoc.2025.113942}, journal = {Applied Soft Computing}, month = {12}, pages = {113942}, shortjournal = {Appl. Soft. Comput.}, title = {DT and LLM driven intelligent maintenance system for L-DED and DAG-based LLM fault diagnosis evaluation framework}, volume = {185}, year = {2025}, }
• (2025). A new large model with multi-scale feature fusion for fault diagnosis based on unified time series model. ASOC, 185, 113941. (www)

  @article{ , abstract = {Large model technology exemplified by large language models has been applied in the field of industrial fault diagnosis. However, existing large models are optimized for specific equipment types and have yet to fully exploit the potential of time-series monitoring data to enable widespread application across diverse mechanical equipment in various industrial scenarios. To address this challenge, a fault diagnosis large model (UniTS-FD) is designed based on unified time series model (UniTS). First, a multi-scale feature fusion backbone network is developed based on UniTS backbone to capture general mechanical fault features. Second, the fault classification head integrates the Pearson correlation coefficient to assess the similarity of class information within linear space for enabling adaptive classification. Third, P-LoRA fine-tuning approach incorporating LoRA and prompt technology is proposed to fine-tune the fault classification head, which enhances the generalization ability of the UniTS-FD model for fault diagnosis tasks of various mechanical equipment. Finally, the UniTS-FD model is pre-trained on 11 fault datasets and fine-tuning experiments were conducted on four different fault datasets to achieve cross-machine fault diagnosis. Experimental results demonstrate the effectiveness of the UniTS-FD in fault diagnosis tasks.}, archive = {J_ASOC}, author = {Zhiwei Zhang and Chengbin Wei and Weimin Zhang and Long Wen}, doi = {10.1016/j.asoc.2025.113941}, journal = {Applied Soft Computing}, month = {12}, pages = {113941}, shortjournal = {Appl. Soft. Comput.}, title = {A new large model with multi-scale feature fusion for fault diagnosis based on unified time series model}, volume = {185}, year = {2025}, }
• (2025). A federated learning-based method for personalized manufacturing service recommendation with collaborative relationships. ASOC, 185, 113940. (www)

  @article{ , abstract = {In the industrial Internet environment, the increasing complexity of manufacturing tasks has rendered them no longer accomplishable by independent manufacturing services. Meanwhile, current recommendation systems predominantly face challenges in maintaining data privacy and security during client parameter exchanges. To address these issues, this paper proposes CoFedSVD+ +, a federated learning-based method for personalized manufacturing service recommendation that integrates an enhanced SVD+ + algorithm with homomorphic encryption. First, we devise an enhanced similarity calculation method to analyze collaborative relationships among manufacturing services. Second, we implement a homomorphic encryption protocol within the federated learning framework to resolve data isolation challenges. Third, the improved SVD+ + algorithm is employed to capture implicit feedback information and predict missing Quality of Service (QoS) metrics. Fourth, a Top-N service composition recommendation list is generated through synergistic analysis of collaborative relationships and QoS predictions. Finally, we validate our approach using real-world case data from an industrial Internet platform. Experimental comparisons with existing recommendation algorithms demonstrate superior recommendation effectiveness of the proposed method.}, archive = {J_ASOC}, author = {Lei Wang and Jun Wang and Feng Xiang and Tongshun Li and Yang Xu and Yibing Li}, doi = {10.1016/j.asoc.2025.113940}, journal = {Applied Soft Computing}, month = {12}, pages = {113940}, shortjournal = {Appl. Soft. Comput.}, title = {A federated learning-based method for personalized manufacturing service recommendation with collaborative relationships}, volume = {185}, year = {2025}, }
• (2025). Span-level emotion-cause-category triplet extraction with instruction tuning LLMs and data augmentation. ASOC, 185, 113938. (www)

  @article{ , abstract = {Span-level emotion-cause-category triplet extraction is a fine-grained task in emotion cause analysis that aims to identify emotion spans, cause spans, and their corresponding emotion categories from documents. Existing methods, including clause-level emotion-cause pair extraction and span-level emotion-cause detection, often suffer from redundant information and difficulties in accurately classifying emotion categories, particularly when emotions are expressed implicitly or ambiguously. To overcome these challenges, this study explores a fine-grained approach to span-level emotion-cause-category triplet extraction and introduces an innovative framework that leverages instruction tuning and data augmentation techniques based on large language models. The proposed method employs task-specific triplet extraction instructions and utilizes low-rank adaptation to fine-tune large language models, eliminating the necessity for intricate task-specific architectures. Furthermore, an LLM-based data augmentation strategy is developed to address data scarcity by guiding large language models in generating high-quality synthetic training data. Extensive experimental evaluations demonstrate that the proposed approach significantly outperforms existing baseline methods, achieving at least a 12.8 % improvement in span-level emotion-cause-category triplet extraction metrics. The results demonstrate the method’s effectiveness and robustness, offering a promising avenue for advancing research in emotion cause analysis.}, archive = {J_ASOC}, author = {Xiangju Li and Dong Yang and Xiaogang Zhu and Faliang Huang and Peng Zhang and Zhongying Zhao}, doi = {10.1016/j.asoc.2025.113938}, journal = {Applied Soft Computing}, month = {12}, pages = {113938}, shortjournal = {Appl. Soft. Comput.}, title = {Span-level emotion-cause-category triplet extraction with instruction tuning LLMs and data augmentation}, volume = {185}, year = {2025}, }
• (2025). Deep cross-visual semantic hashing with self-calibrated collaborative attention. ASOC, 185, 113937. (www)

  @article{ , abstract = {Deep hashing has garnered considerable attention due to its remarkable retrieval efficiency and low storage cost, particularly in visual retrieval scenarios. However, current deep hashing methods generally integrate hash coding into a single-stream architecture, which limits the discriminative power of learned visual features and yields suboptimal hash codes. Additionally, over-reliance on semantic labels shared across samples fails to fully exploit the intrinsic semantic correlations between labels and corresponding visual features. To address these issues, we propose a deep cross-visual semantic hashing (DCvSH) method for image retrieval. First, we develop a visual image feature decoupling encoding network that leverages a self-calibrated collaborative attention mechanism to disentangle common and specific semantics across related images. These decoupled features are fed into a shared decoder for image reconstruction, yielding discriminative visual feature representations. Second, we construct a cross-visual semantic representation learning network with a two-level multi-layer perceptron to capture the underlying relationships between semantic label encodings and visual feature embeddings, while a hypergraph structure is introduced to preserve pairwise similarity relationships. Experimental results on the CIFAR-10, NUS-WIDE, and MIRFLICKR datasets demonstrate consistent improvements, with average mean average precision (mAP) scores reaching 0.895, 0.874, and 0.881 at different code lengths, respectively. Notably, DCvSH outperforms other baselines across all evaluation metrics.}, archive = {J_ASOC}, author = {Hao Feng and Xiangbo Zhou and Yue Wu and Jian Zhou and Banglei Zhao}, doi = {10.1016/j.asoc.2025.113937}, journal = {Applied Soft Computing}, month = {12}, pages = {113937}, shortjournal = {Appl. Soft. Comput.}, title = {Deep cross-visual semantic hashing with self-calibrated collaborative attention}, volume = {185}, year = {2025}, }
• (2025). Large model for fault diagnosis of industrial equipment based on a knowledge graph construction. ASOC, 185, 113936. (www)

  @article{ , abstract = {To address the significant heterogeneity of multi-modal data and the challenges in capturing fault semantics for industrial equipment, a fault diagnosis framework that integrates a time-frequency knowledge graph with the large model DeepSeek-V3 is proposed. Specifically, an unsupervised knowledge graph construction method is designed based on multi-modal vibration data signals. This method mines temporal evolution relationships using dynamic time warping and quantifies the relevance between features and faults via mutual information, thereby forming a dynamic graph representation. Additionally, DeepSeek-V3 encodes the natural language descriptions of vibration features, integrating graph structure and time-frequency map features to achieve collaborative reasoning and diagnosis among text, graphs, and maps. Experimental results show that the proposed method achieves high accuracy and significantly outperforms benchmark models, surpassing traditional methods. The proposed framework, through the deep integration of data-driven knowledge graphs and large model semantic understanding, demonstrates high precision, strong robustness, and transparent decision-making capabilities, providing new insights for intelligent diagnosis of industrial equipment.}, archive = {J_ASOC}, author = {Jichao Zhuang and Jiaming Yang and Weigang Li and Jian Chen and Yunjun Zheng and Zhuyun Chen}, doi = {10.1016/j.asoc.2025.113936}, journal = {Applied Soft Computing}, month = {12}, pages = {113936}, shortjournal = {Appl. Soft. Comput.}, title = {Large model for fault diagnosis of industrial equipment based on a knowledge graph construction}, volume = {185}, year = {2025}, }
• (2025). Interval-valued pythagorean fuzzy distance-based extended inferior ratio method for multiattribute decision-making: Application to green supplier selection in manufacturing industry. ASOC, 185, 113935. (www)

  @article{ , abstract = {Interval-valued Pythagorean fuzzy sets (IVPFSs) have emerged as a powerful tool for handling uncertainty and vagueness in multiattribute decision-making (MADM). In this paper, we first propose a novel distance measure for IVPFSs based on triangular divergence, which satisfies all core distance axioms and significantly improves discrimination ability compared to existing measures. Building on this, we introduce a maximizing deviation strategy with a new loss function to objectively determine attribute weights. Furthermore, we develop an extended inferior ratio (EIR) method that incorporates a dynamic weight parameter to flexibly balance the influence of positive and negative ideal solutions. The performance of the proposed method is demonstrated through a case study on green supplier selection in the manufacturing industry. The results indicate that, among the seven criteria evaluated, the most suitable suppliers are ranked as follows: β (1.0000), α (0.6471), δ (0.3500), ϵ (0.0690), and θ (0.0000). In addition, sensitivity and comparative analyses confirm the robustness and consistency of the proposed method, reflecting its effectiveness and practical value for sustainable decision-making in real-world scenarios.}, archive = {J_ASOC}, author = {Zhe Liu and Donglai Wang and Muhammet Deveci and Sukumar Letchmunan}, doi = {10.1016/j.asoc.2025.113935}, journal = {Applied Soft Computing}, month = {12}, pages = {113935}, shortjournal = {Appl. Soft. Comput.}, title = {Interval-valued pythagorean fuzzy distance-based extended inferior ratio method for multiattribute decision-making: Application to green supplier selection in manufacturing industry}, volume = {185}, year = {2025}, }
• (2025). Multidimensional trade performance assessment for integrating sustainability and economic diversification in OECD countries using a spherical fuzzy SIWEC-SPC-based decision support model. ASOC, 185, 113934. (www)

  @article{ , abstract = {Countries participate in regional trade blocks to enhance their trade activities and trade performance. This study proposes a novel approach to trade performance assessment, introducing innovations both in terms of performance parameters and methodology. From a performance parameter perspective, the inclusion of sustainable development and economic diversity levels in trade performance evaluation differentiates this study from traditional trade performance measurement approaches. From a methodological perspective, a hybrid model that simultaneously considers both subjective and objective criterion weighting based on expert opinions is introduced. The spherical fuzzy (SF)−simple weight calculation (SIWEC)−symmetry point of criterion (SPC)−opportunity losses-based polar coordinate distance (OPLO-POCOD) model, developed as a decision support system, is proposed for evaluating the trade performance of country blocs. SF-SIWEC is employed as the subjective criterion weighting method, while SPC serves as the objective criterion weighting method. The OPLO-POCOD alternative ranking method is employed to calculate the trade performance of countries. The proposed methodology is applied to the Organization for Economic Cooperation and Development (OECD) countries. According to the subjective weighting method, "sustainable development (0.0978)" emerges as the most important criterion, whereas in the objective weighting method, "number of product types (export) (0.1612)" is identified as the most significant one. In the final criterion weights, "number of product types (export) (0.1232)" is also determined to be the most important criterion. Considering the final criterion weights, the United States (0.9871) has the highest trade performance among the OECD countries. Thus, when both sustainability and economic diversification are considered, it is understood that the most influential criteria in determining multidimensional trade performance are the number of exported product types and sustainable development. From this perspective, the United States stands out as the country with the highest multidimensional trade performance among OECD countries. This study contributes to the literature by integrating sustainability and economic diversification parameters into trade performance measurement as well as proposing a comprehensive methodology for performance evaluation in trade blocs.}, archive = {J_ASOC}, author = {Galip Cihan Yalçın and Karahan Kara and Emre Kadir Özekenci and Vladimir Simic and Dragan Pamucar}, doi = {10.1016/j.asoc.2025.113934}, journal = {Applied Soft Computing}, month = {12}, pages = {113934}, shortjournal = {Appl. Soft. Comput.}, title = {Multidimensional trade performance assessment for integrating sustainability and economic diversification in OECD countries using a spherical fuzzy SIWEC-SPC-based decision support model}, volume = {185}, year = {2025}, }
• (2025). Optimizing supply routes in a network interdiction model with dual defense operations. ASOC, 185, 113933. (www)

  @article{ , abstract = {This study introduces a novel bi-level defender–attacker model (BDAM) designed to address real-world homeland defense scenarios. Building on the shortest path network interdiction problem (SPNIP), BDAM incorporates dual defense operations—node interdiction and edge destruction—while explicitly modeling the defender’s supply support. Unlike conventional SPNIP formulations, BDAM jointly considers the attacker’s path disruption and the defender’s logistical requirements, ensuring that all interdicted nodes are supported by available supply nodes without exceeding their capacity. To solve this NP-hard problem, a hybrid metaheuristic algorithm named Improved Simplified Swarm Optimization with Dijkstra (iSSOD) is proposed. The method integrates a population-based SSO framework with a randomized repair mechanism to ensure feasibility and an entropy-guided local search to enhance exploitation. The attacker’s optimal response is computed efficiently using Dijkstra’s algorithm, embedded within the defender’s fitness evaluation. The experimental results on 36 artificial instances demonstrate that iSSOD consistently outperforms several benchmark evolutionary algorithms, providing high-quality solutions through a defense-aware, supply-constrained optimization framework. Furthermore, a real-world case study based on geographic data validates the model’s applicability under realistic defense conditions.}, archive = {J_ASOC}, author = {Wei-Chang Yeh and Chyh-Ming Lai and Tsung-Hua Wu}, doi = {10.1016/j.asoc.2025.113933}, journal = {Applied Soft Computing}, month = {12}, pages = {113933}, shortjournal = {Appl. Soft. Comput.}, title = {Optimizing supply routes in a network interdiction model with dual defense operations}, volume = {185}, year = {2025}, }
• (2025). A continuous encoding-based representation for efficient multi-fidelity multi-objective neural architecture search. ASOC, 185, 113932. (www)

  @article{ , abstract = {Neural architecture search (NAS) is a powerful tool for automatically designing optimized deep learning models but is often limited by high computational costs, especially in multi-objective settings. To address this, an adaptive Co-Kriging-assisted multi-fidelity multi-objective NAS algorithm is proposed to reduce the computational cost and accelerate convergence of NAS by incorporating a clustering-based local multi-fidelity infill sampling strategy. Additionally, we introduce a novel continuous encoding method to compactly represent node connections within a generalized U-Net backbone, significantly reducing search dimensionality. Extensive experiments demonstrate that our method consistently outperforms state-of-the-art NAS approaches under limited computational budgets on three benchmarks, a 2D Darcy flow regression task, a CHASE_DB1 biomedical image segmentation task, and an urban wind velocity prediction task. Analysis further shows that our algorithm autonomously identifies design patterns consistent with expert-curated U-Net variants in literature, confirming its efficiency and potential for insight into performant architectures.}, archive = {J_ASOC}, author = {Zhao Wei and Chin Chun Ooi and Yew-Soon Ong}, doi = {10.1016/j.asoc.2025.113932}, journal = {Applied Soft Computing}, month = {12}, pages = {113932}, shortjournal = {Appl. Soft. Comput.}, title = {A continuous encoding-based representation for efficient multi-fidelity multi-objective neural architecture search}, volume = {185}, year = {2025}, }
• (2025). Meta-learning-based adaptive operator selection for traveling salesman problem. ASOC, 185, 113930. (www)

  @article{ , abstract = {In evolutionary optimization, effectively leveraging knowledge about search operator performance is crucial for enhancing algorithmic results. Traditional operator selection strategies often rely on fixed heuristics or trial-and-error, which struggle to adapt to the nonstationary search dynamics of evolutionary runs—i.e., the stage-dependent, instance-dependent, and population-dependent shifts in operator effectiveness—and typically yield suboptimal performance. To address these challenges, we propose a novel meta-learning-based adaptive operator selection (AOS) framework. It leverages a Long Short-Term Memory (LSTM) neural network to learn temporal patterns of operator performance from historical data and dynamically adjust operator choice on-the-fly. The framework also integrates domain-specific biases to preserve population diversity and promote effective exploration, and it continuously updates its selection policy through dynamic online learning as the evolutionary process unfolds. Experiments on the Traveling Salesman Problem (TSP) benchmark demonstrate that the proposed LSTM-based AOS method significantly outperforms conventional approaches to operator selection. In particular, it achieved a median optimality gap of 9.87 % on a suite of TSP instances—approximately a 20 % improvement over the best fixed-operator configuration—indicating superior solution quality. Moreover, our approach consistently surpassed other state-of-the-art AOS techniques, underscoring the efficacy of the LSTM-driven framework and its significant potential to enhance evolutionary algorithm performance on complex optimization tasks.}, archive = {J_ASOC}, author = {Ho Young Jeong and Byung Duk Song}, doi = {10.1016/j.asoc.2025.113930}, journal = {Applied Soft Computing}, month = {12}, pages = {113930}, shortjournal = {Appl. Soft. Comput.}, title = {Meta-learning-based adaptive operator selection for traveling salesman problem}, volume = {185}, year = {2025}, }
• (2025). A fast and robust ensemble evolving pixel cloud-based image segmentation approach. ASOC, 185, 113926. (www)

  @article{ , abstract = {The existing cluster-based image segmentation algorithms have the burden of iterative computation caused by the change of cluster centers and are sensitive to noise. In this paper, we present a fast and robust ensemble evolving pixel cloud-based image segmentation approach. The concept of pixel clouds by clustering pixels of the same pattern around their focal pixels is proposed. The following attributes distinguish the proposed algorithm: (1) The pixel clouds are evolvable according to the global densities of the incoming pixels and the number of pixel clouds is automatically determined. (2) The focal pixels of pixel clouds are dynamically updated with the highest local densities by using the recursive density estimation, which avoids redundant distance calculations when a new pixel arrives. (3) A multiscale morphological gradient reconstruction operation is employed to merge or filter meaningless pixel clouds, especially in noisy images, which helps to adaptively polish neighboring pixel clouds and compact the pixel clouds. (4) An ensemble structure is introduced to fasten the image segmentation speed by splitting the whole image into multiple independent sub-images, in which the pixel clouds are independently formed and evolved. Comprehensive experiments on natural images, remote sensing images and medical images reveal that the proposed approach surpasses the state-of-the-art algorithms in both segmentation accuracy and computational efficiency. Even for the noisy images, the proposed approach demonstrates more robust performance.}, archive = {J_ASOC}, author = {Tao Zhang and Hai-Jun Rong and Zhao-Xu Yang and Chi-Man Vong}, doi = {10.1016/j.asoc.2025.113926}, journal = {Applied Soft Computing}, month = {12}, pages = {113926}, shortjournal = {Appl. Soft. Comput.}, title = {A fast and robust ensemble evolving pixel cloud-based image segmentation approach}, volume = {185}, year = {2025}, }
• (2025). A data-driven approach to tackling academic stress-coping and mental health issues in college students using spherical fuzzy MARCOS methodology. ASOC, 185, 113925. (www)

  @article{ , abstract = {The drastically developing nature of the knowledge economy and the rising need for top-notch expertise have placed tremendous pressure on college students. As higher education becomes more accessible, masses of students are enrolling in colleges, which puts additional pressure on colleges and institutions; as a result, they cannot provide adequate resources to the students. As the class size increases, many students require mental health assistance, academic guidance, and financial aid, which then puts pressure on the teachers and the facilities. This flood of students overloads the facilities, resulting in it becoming more challenging to provide attention and concern, leading many students to feel overlooked and affecting their mental health. Due to not getting timely support, students may find it challenging to handle their academic responsibilities. Moreover, the students face a heavy workload, unclear guidance, and limited resource access. The objective of this study is to develop a structured, data-driven decision-making framework for systematically evaluating and improving student mental health and academic stress-coping strategies in a college setting. To address this, a comprehensive decision-making structure, measurement of alternatives, and ranking according to the compromised solution (MARCOS) within the spherical fuzzy (SF) environment, has been applied, which evaluates the key factors causing mental health issues by comparing the ideal and anti-ideal alternatives. The novelty of the proposed approach lies in leveraging the SF framework’s explicit ability to model hesitation (abstinence) alongside truth and falsity degrees, enabling more accurate representation of subjective psychological assessments compared to traditional fuzzy models. Furthermore, the method calculates utility functions corresponding to each alternative (coping technique), prioritizes the strategies, and selects the most effective intervention. The results reveal that personalized mental health plans emerged as the top-ranked coping strategy, highlighting the importance of tailored support in culturally and contextually diverse academic environments.}, archive = {J_ASOC}, author = {Raiha Imran and Munazza Amin and Kifayat Ullah and Dragan Pamucar and Zeeshan Ali and Oumaima Saidani and Vladimir Simic}, doi = {10.1016/j.asoc.2025.113925}, journal = {Applied Soft Computing}, month = {12}, pages = {113925}, shortjournal = {Appl. Soft. Comput.}, title = {A data-driven approach to tackling academic stress-coping and mental health issues in college students using spherical fuzzy MARCOS methodology}, volume = {185}, year = {2025}, }
• (2025). Secure and efficient vehicle control of autonomous vehicles using federated deep reinforcement learning. ASOC, 185, 113924. (www)

  @article{ , abstract = {Autonomous driving is largely considered a revolutionary technology and the ultimate alternative for smart urban mobility. Artificial Intelligence (AI) has been the main pillar of AV technology. However, despite the significant success achieved through the application of various Deep Reinforcement Learning (DRL) techniques in the field of AVs, numerous challenges remain to be addressed. Particularly, in the case of vehicle control, existing works have shown limited results when extensive action and state spaces are involved, and where the efficacy of agents is often compromised due to the variability of states and actions. In response to these challenges, we introduce a pioneering approach leveraging Federated Deep Reinforcement Learning (FDRL), which fosters the exchange of experiences among participating agents while safeguarding their privacy. FDRL enhances the exploration of all agents’ actions and states in different environments and effectively overcomes the problem of low sample efficiency in action and state spaces for all agents. Our approach is based on three distinct FDRL algorithms: Federated Proximal Policy Optimization (FPPO), Federated Deep Deterministic Policy Gradient (FDDPG), and Federated Deep Q-Network (FDQN), each tailored to control AVs within unique environments. Our results demonstrate the superiority of our approach in terms of average reward, waiting time, and speed, when compared to local models such as PPO, DDPG, and DQN.}, archive = {J_ASOC}, author = {Badr Ben Elallid and Nabil Benamar and Miloud Bagaa and Nabil Mrani}, doi = {10.1016/j.asoc.2025.113924}, journal = {Applied Soft Computing}, month = {12}, pages = {113924}, shortjournal = {Appl. Soft. Comput.}, title = {Secure and efficient vehicle control of autonomous vehicles using federated deep reinforcement learning}, volume = {185}, year = {2025}, }
• (2025). Simulating phenomenal consciousness using generative agents based on large language models. ASOC, 185, 113922. (www)

  @article{ , abstract = {Large Language Models (LLMs) still face challenges in tasks requiring understanding implicit instructions and applying common-sense knowledge. In such scenarios, LLMs may require multiple attempts to achieve human-level performance, potentially leading to inaccurate responses or inferences in practical environments, affecting their long-term consistency and behavior. This paper introduces the Internal Time-Consciousness Machine (ITCM), a computational consciousness structure to simulate the process of human consciousness. We further propose the ITCM-based Agent (ITCMA), which supports action generation and reasoning in open-world settings, and can independently complete tasks. ITCMA enhances LLMs’ ability to understand implicit instructions and apply common-sense knowledge by considering agents’ interaction and reasoning with the environment. The trained ITCMA performs better than state-of-the-art (SOTA) in the seen set. Even untrained ITCMA can achieve higher task completion rates than SOTA on the seen set, indicating its superiority over traditional intelligent agents in utility and generalization. In real-world tasks with quadruped robots, the task completion rate of untrained ITCMA is close to its performance in the unseen set, demonstrating its comparable utility and universality in real-world settings. CCS Concepts: ∙ Human-centered computing → Interactive systems and tools; ∙ Computing methodologies → Natural language processing.}, archive = {J_ASOC}, author = {Hanzhong Zhang and Jibin Yin and Haoyang Wang and Ziwei Xiang}, doi = {10.1016/j.asoc.2025.113922}, journal = {Applied Soft Computing}, month = {12}, pages = {113922}, shortjournal = {Appl. Soft. Comput.}, title = {Simulating phenomenal consciousness using generative agents based on large language models}, volume = {185}, year = {2025}, }
• (2025). Asymmetrical variable driven simulated binary crossover operator for large scale multi-objective optimization. ASOC, 185, 113921. (www)

  @article{ , abstract = {Due to the excessive number of decision variables, large scale multi-objective optimization has gradually become a hot research topic over the past few years. Although various reproduction operators have been proposed to effectively generate offspring solutions in the huge decision space, they have a common issue in that most of them are focused on the symmetrical variables and neglect asymmetrical variables. Based on the summary above, it is empirically proved that asymmetrical variables with great similarities play an important role in facilitating each other ′ s effective exploration of the huge decision space. Therefore, to effectively tackle large scale multi-objective optimization problems, the asymmetrical variables are incorporated into the design of a new asymmetrical variable driven simulated binary crossover operator. On the basis of the proposed asymmetrical variable driven simulated binary crossover operator, this paper further constructs a large scale multi-objective evolutionary framework based on asymmetrical variable driven simulated binary crossover operator. Extensive experiments and analyses on typical large scale multi-objective optimization problems with up to 10,000 decision variables and practical engineering problems with 6000 decision variables show its superiority over state-of-the-art optimizers.}, archive = {J_ASOC}, author = {Maoqing Zhang and Kun Wang}, doi = {10.1016/j.asoc.2025.113921}, journal = {Applied Soft Computing}, month = {12}, pages = {113921}, shortjournal = {Appl. Soft. Comput.}, title = {Asymmetrical variable driven simulated binary crossover operator for large scale multi-objective optimization}, volume = {185}, year = {2025}, }
• (2025). Towards agentic smart design: An industrial large model-driven human-in-the-loop agentic workflow for geometric modelling. ASOC, 185, 113920. (www)

  @article{ , abstract = {Agentic workflows, powered by Industrial Large Models (ILMs), represent a significant development emphasizing collaboration between humans and intelligent systems. This paper presents a structured perspective on the role of agentic workflows for smart design and manufacturing, grounded in integrating ILMs. We define an agentic workflow as a labeled Activity-on-Vertex (AOV) graph, where each node represents a functionally closed subtask and is executed by an ILM-based agent, a human operator, or an automated system. This formalism supports analyzable, modular, and hybrid execution, offering a foundation for modeling complex, mixed-initiative processes in manufacturing environments. To support real-world deployment, we introduce a set of reusable agentic workflow patterns that describe how ILM agents perceive, plan, and act in coordination with other components. Besides, a proof-of-concept case study illustrates the practical application of the human-in-the-loop framework through the agentic generation of CAD models. The study covers task decomposition, workflow implementation, and benchmarking, providing evidence for the feasibility of agentic workflows. Building upon these findings, this work contributes to advancing the development and application of agentic workflows in smart manufacturing contexts.}, archive = {J_ASOC}, author = {Keyou Zheng and Yuanwei Zhong and Xuyang Su and Jiewu Leng and Qiang Liu and Xin Chen}, doi = {10.1016/j.asoc.2025.113920}, journal = {Applied Soft Computing}, month = {12}, pages = {113920}, shortjournal = {Appl. Soft. Comput.}, title = {Towards agentic smart design: An industrial large model-driven human-in-the-loop agentic workflow for geometric modelling}, volume = {185}, year = {2025}, }
• (2025). Adaptive fault diagnosis of railway vehicle on-board controller with large language models. ASOC, 185, 113919. (www)

  @article{ , abstract = {Accurately identifying fault types of the Vehicle On-Board Controller (VOBC) in railway systems is of great significance for ensuring the safe operation of trains. Recently, Large Language Models (LLMs) have demonstrated excellent semantic understanding and natural language interaction capabilities, offering a novel solution for VOBC fault diagnosis. However, LLMs pretrained on general domains lack specific knowledge related to railway VOBC fault diagnosis scenarios, resulting in insufficient adaptability to railway-specific text corpora. This paper conducts an in-depth study on the adaptability of LLMs to VOBC fault diagnosis and proposes Railway Fault Diagnosis Large Language Model (RFD-LLM). First, we adopt railway domain adaptation based on Low-Rank Adaptation (LoRA) to match VOBC fault patterns. Second, instruction tuning is applied to achieve domain knowledge alignment and enhance the model’s ability to follow instructions. The proposed RFD-LLM is the first large language model-based fault diagnosis model for railway VOBC, capable of efficiently and accurately identifying seven types of VOBC fault patterns. RFD-LLM provides a new solution for the development of large models in the railway domain.}, archive = {J_ASOC}, author = {Cong Peng and Jiali Peng and Zisheng Wang and Zongyao Wang and Junjie Chen and Jianping Xuan and Tielin Shi}, doi = {10.1016/j.asoc.2025.113919}, journal = {Applied Soft Computing}, month = {12}, pages = {113919}, shortjournal = {Appl. Soft. Comput.}, title = {Adaptive fault diagnosis of railway vehicle on-board controller with large language models}, volume = {185}, year = {2025}, }
• (2025). An MADM model using frank operations based power aggregation operator under p,q-quasirung orthopair fuzzy sets for highway selection in war-plane landing. ASOC, 185, 113918. (www)

  @article{ , abstract = {In military logistics and operational planning, selecting an optimal highway for war-plane landings and take-offs is a critical and strategic decision. This process involves several key factors that directly affect mission success, operational safety, and public security. Among the most important attributes are the highway’s straight and long stretch with sufficient width to accommodate war-plane landing distances, and its surface condition, which must be free from obstacles, debris, and damage. Low traffic density is crucial to avoid the risk of collisions during landing. Additionally, favourable weather conditions, proximity to military camps, availability of emergency services and fuel, and a secure and hazard-free surrounding terrain are essential for safe and efficient operations. These factors collectively form the backbone of a reliable and tactical approach to highway selection for military air operations. Thus, in order to assess and rank various options for the landing and take-off of war planes, a strong and trustworthy procedure for making decisions is required. The purpose of this experiment is to build a comprehensive structure in multi-attribute decision making environment, using suggested p , q -quasirung orthopair fuzzy Frank power averaging as well as p , q -quasirung orthopair fuzzy Frank power geometric operators to capture ambiguity and uncertainty in highway selection. Furthermore, p , q -quasirung orthopair fuzzy Frank power weighted aggregation along with p , q -quasirung orthopair fuzzy Frank power weighted geometric operators are implemented for integrating the distance as well as similarity measures. Finally, sensitivity analysis and a comparison with the present technique are included to further demonstrate the superiority and validity of the technique that is suggested.}, archive = {J_ASOC}, author = {Sanjita Giri and Sankar Kumar Roy and Muhammet Deveci}, doi = {10.1016/j.asoc.2025.113918}, journal = {Applied Soft Computing}, month = {12}, pages = {113918}, shortjournal = {Appl. Soft. Comput.}, title = {An MADM model using frank operations based power aggregation operator under p,q-quasirung orthopair fuzzy sets for highway selection in war-plane landing}, volume = {185}, year = {2025}, }
• (2025). Real-time traffic flow optimization using large language models and reinforcement learning for smart urban mobility. ASOC, 185, 113917. (www)

  @article{ , abstract = {The exponential growth of metropolitan populations causes transportation network congestion, which increases fuel usage, travel time, and environmental damage. Traditional traffic management systems (TMS) seldom handle these issues in real time. Recently developed Large Language Models (LLMs), especially those using Reinforcement Learning (RL), may enhance urban transportation systems. Traffic management technology's real-time flexibility and shifting congestion patterns provide improved potential. Traditional approaches cannot estimate traffic flow or adapt to urban settings. A strong AI-driven method is needed to improve urban mobility and traffic flow. This paper introduces the LLM-RL Traffic Optimization Framework (LLM-RL-TOF). LLMs analyze real-time traffic data and give predictive insights in this context. Due to these new insights, the RL algorithm can improve traffic flow in real time and reduce congestion via dynamic traffic management. IoT sensors and urban traffic cameras capture real-time traffic data, including traffic volume and incidents. This data helps the LLM estimate bottlenecks, accidents, and traffic congestion. An RL agent uses LLM outputs to adjust traffic signal timing and suggest alternate routes. With real-time alternatives, traffic flow and urban mobility may be optimized. The junction throughput rate rose 17.5 %, the queue length accumulation index fell 22.3 %, and the average vehicle delay fell 18.6 %. The decrease in average vehicle delay enabled all these gains.}, archive = {J_ASOC}, author = {Arvind R. Singh and Muhammad Wasim Abbas Ashraf and Rajkumar Singh Rathore and Bin Li and M.S. Sujatha}, doi = {10.1016/j.asoc.2025.113917}, journal = {Applied Soft Computing}, month = {12}, pages = {113917}, shortjournal = {Appl. Soft. Comput.}, title = {Real-time traffic flow optimization using large language models and reinforcement learning for smart urban mobility}, volume = {185}, year = {2025}, }
• (2025). Improving speech emotion recognition using gated cross-modal attention and multimodal homogeneous feature discrepancy learning. ASOC, 185, 113915. (www)

  @article{ , abstract = {Speech emotion recognition (SER) remains a significant and crucial challenge due to the complex and multifaceted nature of human emotions. To tackle this challenge, researchers strive to integrate information from diverse modalities through multimodal learning. However, existing multimodal fusion techniques often overlook the intricacies of interactions between different modalities, resulting in suboptimal feature representations. In this paper, we propose WavFusion, a multimodal framework designed for SER that tackles key research challenges, such as effective multimodal fusion, modality heterogeneity, and discriminative representation learning. By utilizing a gated cross-modal attention mechanism and multimodal homogeneous feature discrepancy learning, WavFusion outperforms existing state-of-the-art methods on benchmark datasets. Our research highlights the importance of capturing subtle cross-modal interactions and learning discriminative representations for accurate multimodal SER. Experimental results indicate that the proposed method is highly competitive and better than most of the latest state-of-the-art methods for SER. WavFusion achieves 0.78 % and 1.27 % improvement in accuracy and 0.74 % and 0.44 % improvement in weighted F1 score over the previous methods on the IEMOCAP and MELD datasets, respectively.}, archive = {J_ASOC}, author = {Feng Li and Jiusong Luo and Wanjun Xia}, doi = {10.1016/j.asoc.2025.113915}, journal = {Applied Soft Computing}, month = {12}, pages = {113915}, shortjournal = {Appl. Soft. Comput.}, title = {Improving speech emotion recognition using gated cross-modal attention and multimodal homogeneous feature discrepancy learning}, volume = {185}, year = {2025}, }
• (2025). Wind turbine blades defect detection based on global and local attention with multi-feature fusion. ASOC, 185, 113914. (www)

  @article{ , abstract = {Wind turbine blades are prone to small-scale defects—such as cracks, corrosion, and contamination—during long-term operation. Accurate detection of these defects is essential for ensuring the safety and efficiency of wind power systems. However, small-object detection remains challenging due to limited feature representation and weak discriminative cues. To address this, an enhanced YOLOX-s-based framework called Global-Frequency Dual-aware YOLOX (GFD-YOLOX) is proposed. GFD-YOLOX introduces three main improvements. First, the Path Aggregation Feature Pyramid Network (PAFPN) in the neck is replaced with Dual-Frequency Fused Bidirectional Feature Pyramid Network (DFF-BiFPN) to strengthen multi-scale contextual representation. Second, the backbone bottleneck is redesigned with a lightweight structure, improving computational efficiency and convergence speed. Third, a Hierarchical Frequency-Adaptive Fusion (HFAF) module is integrated to enhance cross-scale feature interaction by combining fine-grained and global information. On the self-constructed WTBlade-Defect dataset (3570 annotated images, five defect types: corrosion, hide-craze, surface-eye, thunderstrike, dirt), GFD-YOLOX achieves mAP@0.5 and mAP@0.5:0.95 scores of 94.5 % and 68.9 %, respectively, with 44.3 FPS inference—improving by 13.6 % and 14.4 % over state-of-the-art models. On the public dataset of Ashley Foster et al., it achieves 94.8 % and 69.3 %, with gains of 10.4 % and 10.9 %. These results demonstrate that GFD-YOLOX delivers substantial accuracy gains while maintaining real-time speed and strong cross-dataset generalization, indicating high potential for deployment in operational wind turbine inspection systems.}, archive = {J_ASOC}, author = {Dandan Liu and Mingjie Liu}, doi = {10.1016/j.asoc.2025.113914}, journal = {Applied Soft Computing}, month = {12}, pages = {113914}, shortjournal = {Appl. Soft. Comput.}, title = {Wind turbine blades defect detection based on global and local attention with multi-feature fusion}, volume = {185}, year = {2025}, }
• (2025). Heterogeneous feature selection with group structure mining in fuzzy decision systems for medical diagnosis. ASOC, 185, 113913. (www)

  @article{ , abstract = {In practical applications such as medical diagnosis and group decision making, the potential structural information contained in multi-dimensional features in the form of group domains plays an important role. However, most existing feature selection methods adopt transformed feature spaces for group structure analysis, which lack intrinsic semantic information interpretation. Meanwhile, fuzzy and uncertain heterogeneous data acquired from multiple devices increase the difficulty of task learning. Motivated by these two issues, this work devises a Heterogeneous Feature Selection method with Group Structure Mining in fuzzy decision systems (HFS-GSM), which follows the principle of one “strategy" and one “mechanism". Specifically, a feature group generation strategy based on fuzzy approximation Markov blanket is first designed for mining features with group structure, which introduces the concept of Markov blanket into the fuzzy rough set and utilizes the idea of approximation and fuzzy uncertainty measures. Then, a fuzzy dependency-based overlapping group elimination mechanism is proposed by attribution division, which avoids local redundancy while preserving global discriminative information. Furthermore, the effectiveness of HFS-GSM is verified in comparison with seven representative feature selection methods on publicly available medical datasets. Finally, medical diagnosis data provided by a hospital are obtained to demonstrate the reliability and utility of HFS-GSM in practical applications.}, archive = {J_ASOC}, author = {Jihong Wan and Hongmei Chen and Li Xiao and Chuangpeng Shen and Wei Huang and Xiaoping Li}, doi = {10.1016/j.asoc.2025.113913}, journal = {Applied Soft Computing}, month = {12}, pages = {113913}, shortjournal = {Appl. Soft. Comput.}, title = {Heterogeneous feature selection with group structure mining in fuzzy decision systems for medical diagnosis}, volume = {185}, year = {2025}, }
• (2025). Ahybrid ICEEMDAN-BO-GRU model for real-time grouting parameter prediction. ASOC, 185, 113912. (www)

  @article{ , abstract = {Unit injection volume is one of the core control indicators for evaluating the construction quality of curtain grouting. In actual construction processes, accurate prediction of unit injection volume enables early warning of abnormal conditions during construction. However, geological conditions and construction parameters significantly influence unit injection volume, among other factors, and the data exhibits high nonlinearity. Previous studies have achieved suboptimal prediction accuracy. To address the issue of low prediction accuracy in traditional algorithms when handling highly nonlinear data, this study first applies the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) algorithm to decompose the raw data, and the nonlinear features in the curtain grouting data are effectively eliminated. Subsequently, the Gated Recurrent Unit (GRU) and Bayesian Optimization (BO) algorithms are organically integrated, and Bayesian optimization is applied to the hyperparameters of the GRU model to enhance its predictive performance, thereby constructing an intelligent predictive model for the curtain grouting unit injection volume. Finally, an engineering application study was conducted using an actual curtain grouting project at a pumped-storage power station. The results indicate that the model exhibits good computational accuracy and efficiency, providing theoretical and methodological support for constructing intelligent prediction models in curtain grouting.}, archive = {J_ASOC}, author = {Fei Tong and Dongqing Bai and Xufei Ma and Lin Cheng and Jie Yang and Xiangyu Cao}, doi = {10.1016/j.asoc.2025.113912}, journal = {Applied Soft Computing}, month = {12}, pages = {113912}, shortjournal = {Appl. Soft. Comput.}, title = {Ahybrid ICEEMDAN-BO-GRU model for real-time grouting parameter prediction}, volume = {185}, year = {2025}, }
• (2025). DefGCL: Defence-enhanced graph contrastive learning against attribute inference attacks. ASOC, 185, 113911. (www)

  @article{ , abstract = {Graph-structured data are prevalent in many real-world applications, such as social networks, drug discovery, and fraud detection. While Graph Neural Networks (GNNs) have shown remarkable performance by capturing rich relational patterns, their success often relies on large labeled datasets and raises growing privacy concerns. Graph Contrastive Learning (GCL) has emerged as a powerful unsupervised alternative by leveraging data augmentations to learn robust representations without labeled data. However, recent studies reveal that GCL models are particularly vulnerable to attribute inference attacks, and existing works prioritize performance improvement over privacy protection. To address this issue, we propose a D e f ense-enhanced G raph C ontrastive L earning, dubbed DefGCL , that integrates four coordinated defense strategies to enhance privacy without degrading utility. Specifically, DefGCL employs edge-based graph augmentations to limit exposure to structural attributes, selects negative samples with low attribute sensitivity scores to reduce leakage, modifies the contrastive loss to decouple graph embeddings from attributes, and injects differential privacy noise during the embedding stage. Extensive experiments on five benchmark datasets demonstrate that DefGCL achieves state-of-the-art (SOTA) performance in both privacy preservation and task accuracy. For instance, on the AIDS dataset, DefGCL reduces attribute inference accuracy by 35 % while incurring only a 0.60 % drop in main task performance. Additionally, DefGCL improves computational efficiency by reducing runtime by nearly 50 % compared to baseline methods. 1}, archive = {J_ASOC}, author = {Jinyin Chen and Fanyu Ao and Wenbo Mu and Haiyang Xiong}, doi = {10.1016/j.asoc.2025.113911}, journal = {Applied Soft Computing}, month = {12}, pages = {113911}, shortjournal = {Appl. Soft. Comput.}, title = {DefGCL: Defence-enhanced graph contrastive learning against attribute inference attacks}, volume = {185}, year = {2025}, }
• (2025). Interval multi-feature sparse transformer-CNN: A synergistic approach to precise and efficient spatio-temporal wind speed interval-value prediction. ASOC, 185, 113910. (www)

  @article{ , abstract = {Accurate and reliable wind speed prediction is critical for stabilizing wind power integration. However, current methods are limited by accuracy and stability issues, hindering their large-scale application in wind farms. To overcome these problems, this study innovatively proposes an IMFSformer-CNN model integrating three core components. First, the spatio-temporal and multi-factor feature extraction technology comprehensively captures the spatio-temporal patterns and complex dependencies of wind speed dynamics, incorporating multiple factors such as meteorological variables and spatial correlation. Second, the multi-feature sparse attention mechanism reduces computational complexity by combining sparse attention with multi-feature attention, enhancing representation ability and scalability for precise interval value prediction. Finally, the enhanced interval spatio-temporal prediction fusion model combines the global dependency modeling capabilities of the improved Transformer architecture with the local receptive field advantages of CNN. This hybrid design facilitates the simultaneous capture of both macro-scale atmospheric patterns and micro-scale wind speed fluctuations. The model achieved prediction interval coverage probabilities of 0.921 and 0.899, and coverage width criteria of 1.493 and 3.776, outperforming other models on both datasets. This significantly enhances accuracy and practical value for wind farm cluster forecasting, supporting more reliable and efficient wind energy integration into power grids.}, archive = {J_ASOC}, author = {Weiyi Jiang and Jujie Wang and Xuecheng He}, doi = {10.1016/j.asoc.2025.113910}, journal = {Applied Soft Computing}, month = {12}, pages = {113910}, shortjournal = {Appl. Soft. Comput.}, title = {Interval multi-feature sparse transformer-CNN: A synergistic approach to precise and efficient spatio-temporal wind speed interval-value prediction}, volume = {185}, year = {2025}, }
• (2025). Multi-granularity spatiotemporal fusion neural network with denoising reconstruction for human motion prediction. ASOC, 185, 113909. (www)

  @article{ , abstract = {Human motion prediction has significant application value in many fields. However, existing methods often fail to fully capture the spatial relationships between joints and the temporal flow of information when modeling complex spatiotemporal dependencies. Additionally, these methods are prone to overfitting dominant features while neglecting other important aspects, and struggle with perceiving contour features effectively. To address these issues, this study introduces a novel encoder-decoder framework. The encoder generates a dual-layer adaptive adjacency matrix using a distance partition strategy to parameterize joint relationships, while incorporating a gating mechanism to control the temporal flow of information. The decoder then employs separate spatiotemporal attention modules to decode temporal and spatial features independently. These features are subsequently reconstructed through a spatiotemporal fusion strategy, effectively decoupling and modeling complex spatiotemporal dependencies. To address the issue of overfitting to dominant features, we introduce a denoising reconstruction strategy that allows the model to learn richer combinations of spatiotemporal features under multiple constraints. Furthermore, a multi-granularity information adaptive fusion module is incorporated to achieve adaptive fusion of both local and contour features. Experimental results across several benchmark datasets demonstrate that our method significantly outperforms the state-of-the-art approaches, showcasing its effectiveness in human motion prediction tasks.}, archive = {J_ASOC}, author = {Yong Li and Linfeng Zhu and Haofei Xie and Xinchang Yi}, doi = {10.1016/j.asoc.2025.113909}, journal = {Applied Soft Computing}, month = {12}, pages = {113909}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-granularity spatiotemporal fusion neural network with denoising reconstruction for human motion prediction}, volume = {185}, year = {2025}, }
• (2025). FFNN: Fractional order basis function multi-step neural network method for fractional partial differential equations. ASOC, 185, 113907. (www)

  @article{ , abstract = {With the advancement in artificial intelligence technology, the increasing number of researchers utilize it to address complex equations in ocean engineering. So the technology of artificial intelligence has become a practical area of research. In this paper, we design a novel method to solve the fractional order long water wave equation, which is called the fractional order basis function multi-step neural network. Firstly, a power series is constructed based on a fractional order basis function, which serves as the approximate solution. Secondly, neural networks and the initial conditions of differential equations are integrated into the construction of approximate solutions. Furthermore, the solution is discretized, and a multi-step unfolding strategy is employed on the resulting discrete solution. This approach ensures that each point in the solution is influenced by its predecessor. By means of repeated applications of the optimization algorithm, the residuals are successively diminished, thereby yielding approximate solutions to the equations. Finally, the efficacy and versatility of the proposed strategy were validated through a series of numerical experiments. Compared with the method of fractional physics-informed neural networks, there are up to 18.7 -fold and 22.8 -fold increases in stability of average and maximum residuals. Simultaneously, initial conditions are retained in new solutions.}, archive = {J_ASOC}, author = {Jianke Zhang and Xudong Tian and Chang Zhou}, doi = {10.1016/j.asoc.2025.113907}, journal = {Applied Soft Computing}, month = {12}, pages = {113907}, shortjournal = {Appl. Soft. Comput.}, title = {FFNN: Fractional order basis function multi-step neural network method for fractional partial differential equations}, volume = {185}, year = {2025}, }
• (2025). Discrimination index based attribute reduction for partially labeled heterogeneous data via a new prediction label strategy and statistical distribution of data. ASOC, 185, 113904. (www)

  @article{ , abstract = {Existing methods often struggle to effectively utilize the statistical distribution of data and unlabeled information when dealing with partially labeled heterogeneous data, leading to limited performance in attribute reduction. To address this issue, this paper presents a novel semi-supervised attribute reduction method that integrates statistical distribution of data, conditional discrimination index and a new prediction label strategy. Initially, the similarity between objects is constructed by analyzing the statistical distribution of data. Subsequently, a new prediction label strategy is introduced, which not only utilizes existing label information but also completes missing information by predicting new labels, thereby enhancing the completeness and usability of data. Finally, conditional discrimination index, a tool for measuring the importance of attributes in classification tasks, is employed to design an attribute reduction algorithm. The experimental results on real-world partially labeled heterogeneous datasets show that the proposed p-CDIAR algorithm exhibits superior performance compared to the existing nine reduction algorithms, with an average increase of 4.67 % in classification accuracy and 6.32 % in outlier detection performance.}, archive = {J_ASOC}, author = {Qingnian Li and Haixin Huang and Tao Lu and Huaming Wei and Zhaowen Li}, doi = {10.1016/j.asoc.2025.113904}, journal = {Applied Soft Computing}, month = {12}, pages = {113904}, shortjournal = {Appl. Soft. Comput.}, title = {Discrimination index based attribute reduction for partially labeled heterogeneous data via a new prediction label strategy and statistical distribution of data}, volume = {185}, year = {2025}, }
• (2025). Extended belief rule-based system with online joint learning strategy. ASOC, 185, 113901. (www)

  @article{ , abstract = {With the proliferation of dynamic data streams in the advanced technology environment, it is necessary to solve the evolving classification problems by using adaptable and interpretable artificial intelligence techniques. To meet this challenge, a new extended belief rule-based (EBRB) system incorporating online joint learning strategy is proposed in this paper. The online joint learning strategy comprises two key components: rule update and parameter update schemes. In the rule update scheme, different rule incorporation processes are designed for the labeled or unlabeled input data while overlapping and redundant rules are removed from the rule base. To adapt the updated rule base, the parameter update scheme is designed to retune the parameters within updated rule base. The antecedent attribute weights are optimized using the Bayesian optimization algorithm and the rule weights are updated based on the consistency of rules. To evaluate the performance of the developed system, it is applied to assist radiologists in diagnosing thyroid nodules. Compared with the existing offline EBRB systems and online learning methods, the proposed online joint learning EBRB system could generate higher classification accuracy with fewer rules in the limited running time.}, archive = {J_ASOC}, author = {Bingbing Hou and Min Xue and Leilei Chang and Zijian Wu}, doi = {10.1016/j.asoc.2025.113901}, journal = {Applied Soft Computing}, month = {12}, pages = {113901}, shortjournal = {Appl. Soft. Comput.}, title = {Extended belief rule-based system with online joint learning strategy}, volume = {185}, year = {2025}, }
• (2025). Fusing local density and approximate distance for nonparametric outlier detection. ASOC, 185, 113898. (www)

  @article{ , abstract = {Outlier detection is an essential yet challenging task in intelligent data analysis, and some density-based unsupervised methods have been introduced to identify outliers in low-density regions. However, these methods still suffer from inaccurate density estimation and limited capability in detecting diverse types of outliers. In this study, we propose a nonparametric outlier detection method with the fusion of density and distance (POD-FDD). The proposed method employs adaptive kernel density estimation based on natural neighborhoods, which reduces the sensitivity to parameters in density estimation. Moreover, the optimistic and pessimistic densities are introduced to enhance the reliability of density estimation in the local neighborhood. In addition, approximate reachability distance information is integrated to improve the capability of identifying cluster outliers. Ultimately, a robust parametric-free outlier detection method is developed to detect different types of outliers. Extensive comparative experiments and statistical significance analysis on synthetic and public datasets demonstrate its superior performance, achieving an average improvement of 1.97 % in the AUC metric.}, archive = {J_ASOC}, author = {Zhiyu Chen and Can Gao and Jie Zhou and Ying Yu}, doi = {10.1016/j.asoc.2025.113898}, journal = {Applied Soft Computing}, month = {12}, pages = {113898}, shortjournal = {Appl. Soft. Comput.}, title = {Fusing local density and approximate distance for nonparametric outlier detection}, volume = {185}, year = {2025}, }
• (2025). Prediction of the risk level of saltwater intrusion in the modaomen waterway of the pearl river estuary based on the deep temporal clustering model. ASOC, 185, 113897. (www)

  @article{ , abstract = {In recent years, extreme events caused by global climate change have intensified the phenomenon of saltwater intrusion (SWI) in estuaries. The nonlinear and non-stationary characteristics of estuarine SWI have led to an exponential decline in the timeliness of traditional regression prediction models, making it difficult to meet the operational needs of SWI forecasting. To address this, this study proposed a technical framework for SWI risk level forecasting based on temporal clustering, with its core innovation lying in algorithmic improvements for accurately characterizing complex disaster systems. The key challenges in forecasting SWI risk levels involved capturing the dynamic nonlinear relationships between multidimensional disaster factors (such as runoff, tide level, and wind) and SWI severity, as well as enhancing feature discriminability in label-limited scenarios. Accordingly, this study optimized algorithms through dual-path supervised and unsupervised learning: In the supervised learning framework, LightGBM, RF, XGBoost, and Extra trees were introduced as base learners into the Deep Forest (DF) model. The complementary feature-space partitioning of diverse learners was leveraged to improve the model’s ability to distinguish risk -level boundaries, achieving an average performance gain of 7.8 %. In the unsupervised learning framework, discriminative regularization was incorporated into the Extreme Learning Machine-Autoencoder (ELM-AE) model. By forcing features of samples from the same class to cluster toward the class center, the model’s feature separability for rare events (e.g., severe SWI) was enhanced, leading to an average performance improvement of 11 %. Finally, the optimal model was used to extract dynamic evolution patterns between multidimensional disaster factors and SWI risk levels, with interpretability analysis conducted for real-world forecasting. Notably, upstream flow sequences exhibited high distinguishability between no-SWI and severe-SWI, while mild and moderate SWI showed similar flow patterns, with tidal sequences being the primary differentiator. The algorithmic advancements not only enhanced the accuracy and efficiency of SWI forecasting but also provided a generalizable framework for risk classification in nonlinear hydrological systems.}, archive = {J_ASOC}, author = {Qingqing Tian and Hongyu Yang and Yu Tian and Peiyao Weng}, doi = {10.1016/j.asoc.2025.113897}, journal = {Applied Soft Computing}, month = {12}, pages = {113897}, shortjournal = {Appl. Soft. Comput.}, title = {Prediction of the risk level of saltwater intrusion in the modaomen waterway of the pearl river estuary based on the deep temporal clustering model}, volume = {185}, year = {2025}, }
• (2025). Contrastive learning with mirror-task for multimodal sentiment analysis. ASOC, 185, 113896. (www)

  @article{ , abstract = {Multi-Task Learning (MTL) in Multimodal Sentiment Analysis (MSA) involves implementing various parameter-sharing strategies among tasks. Currently, MSA primarily focuses on hard parameter-sharing mechanisms based on encoder sharing, while soft parameter-sharing is often neglected. To explore a reasonable combination of soft and hard mechanisms in MSA and optimize multimodal representations, along with multimodal contrastive learning, we propose D 3 MSA. It consists of D ouble network (primaryNet and MirrorNet), D ouble parameter-sharing strategies and D ouble contrastive learning modes for multimodal sentiment analysis. D 3 MSA utilizes hard-sharing to consolidate correlations between positive samples of intra-sample contrastive learning. In soft-sharing, we propose a pre-trained MirrorNet (MN) that generates negative samples by the learned inverse distributions. This optimizes the feature space of negative samples. MN interacts with the MSA task through soft-sharing during inter-sample contrastive learning. Experimental results demonstrate that our proposed method can achieve advanced performance on the CMU-MOSI and CMU-MOSEI datasets with lightweight training that requires only a small number of parameters.}, archive = {J_ASOC}, author = {Hang Shi and Lianmin Zhou and Yuanyuan Pu and Zhengpeng Zhao and Jinjing Gu and Dan Xu}, doi = {10.1016/j.asoc.2025.113896}, journal = {Applied Soft Computing}, month = {12}, pages = {113896}, shortjournal = {Appl. Soft. Comput.}, title = {Contrastive learning with mirror-task for multimodal sentiment analysis}, volume = {185}, year = {2025}, }
• (2025). A nearest-best neighbors optimization algorithm for constrained multimodal multiobjective problems. ASOC, 185, 113895. (www)

  @article{ , abstract = {Constrained multimodal multiobjective problems (CMMOPs) have multiple equivalent constrained Pareto optimal sets in the decision space corresponding to the identical constrained Pareto front in the objective space. The key to solving CMMOPs is how to balance feasibility, convergence, and diversity of solutions in both the decision and objective spaces. In view of this, this paper proposes a Nearest-Best neighbors optimization algorithm with constraint-based fitness (NBNOA) to solve CMMOPs. First, a constraint-based fitness assignment scheme is designed to assign specific fitness values to individuals in the population. Then, the Nearest-better-neighbor clustering method is adopted to identify the nearest-better neighbor and best neighbor of each individual according to the specific fitness values. On this basis, a Nearest-Best neighbors guided strategy is developed to guide the search direction of individuals, striking a better balance between exploration and exploitation capabilities. Moreover, a CDP-density elite selection mechanism is constructed to obtain feasible Pareto optimal solutions with higher precision and better diversity. Extensive experiments on two CMMOPs test suites demonstrated that the proposed NBNOA significantly outperforms nine state-of-the-art algorithms. Notably, NBNOA ranks first among all ten algorithms and achieves the best values for 23 out of 31 benchmark functions regarding the reciprocal of Pareto sets proximity and inverted generational distance. Furthermore, NBNOA is applied to a real-world CMMOP, verifying its effective practical application capability. Additionally, NBNOA is tested on two high-dimensional constrained multiobjective optimization problems test suites, further proving its competitive performance in solving complex problems.}, archive = {J_ASOC}, author = {Xuming Han and Ting Zhou and Limin Wang and Yali Chu}, doi = {10.1016/j.asoc.2025.113895}, journal = {Applied Soft Computing}, month = {12}, pages = {113895}, shortjournal = {Appl. Soft. Comput.}, title = {A nearest-best neighbors optimization algorithm for constrained multimodal multiobjective problems}, volume = {185}, year = {2025}, }
• (2025). Frequency decomposition and patch modeling framework for time-series forecasting. ASOC, 185, 113890. (www)

  @article{ , abstract = {Time series forecasting is widely applied across diverse fields, including finance, transportation, and energy, and has made significant contributions in these areas. However, in real-world applications, time series data can be complex and dynamic. Current methodologies still encounter several challenges in managing high-dimensional data, extracting intricate features, and making long-term forecasts. In this study, we propose a Frequency Decomposition and Patch Modeling Framework (FPF). Our FPF consists of the Frequency Domain Decomposition Block (FDB) and the Dual Patch Modeling Block (DPMB). DPMB consists of Patch Enhancement Block and Patch Mixing Block. First, FDB transforms the input sequence to the frequency domain through the Fast Fourier Transform and designs frequency masks to decompose the data into high-frequency and low-frequency components, to extract fast-changing patterns and trend information respectively. Subsequently, DPMB divides the components into patches, where the high-frequency components are modeled by MLP-based Patch Enhancement Block to capture local features, and the low-frequency components are modeled by Transformer-based Patch Mixing Block to capture global dependencies and cross-patch correlations. We conducted comprehensive experiments using seven real-world time series forecasting datasets, including ETT, Traffic, Electricity, and Weather. The findings indicate that this method demonstrates superior performance in the field of time series forecasting.}, archive = {J_ASOC}, author = {Denghui Xu and Hua Wang and Fan Zhang}, doi = {10.1016/j.asoc.2025.113890}, journal = {Applied Soft Computing}, month = {12}, pages = {113890}, shortjournal = {Appl. Soft. Comput.}, title = {Frequency decomposition and patch modeling framework for time-series forecasting}, volume = {185}, year = {2025}, }
• (2025). OdML: An optimization-driven meta-learning framework for multi-task approximate model predictive control. ASOC, 185, 113882. (www)

  @article{ , abstract = {Recent advances in approximate model predictive control (MPC) have leveraged deep neural networks (DNNs) to imitate optimal control laws, significantly reducing the computational cost of real-time optimization. However, these methods often struggle to adapt efficiently to previously unseen system dynamics and suffer from poor data efficiency in real time. In this paper, we address these challenges by treating control problems with different system parameters as separate tasks within a meta-learning scope. Under this scope, we propose OdML, an Optimization-driven Meta-Learning approach for approximate MPC that enables rapid online fine-tuning of DNN-based controllers. We introduce an optimization-driven fine-tuning mechanism that allows for fast adaptation to new tasks using limited online data, without requiring full knowledge of the system parameters. Furthermore, to ensure safety during adaptation, we incorporate control barrier functions into the optimization process, allowing the controller to satisfy safety constraints even under previously unseen conditions. We demonstrate the effectiveness of OdML through various simulation scenarios, highlighting its ability to achieve safe and efficient control.}, archive = {J_ASOC}, author = {Junbo Tong and Shuhan Du and Daming Shi and Wenhui Fan}, doi = {10.1016/j.asoc.2025.113882}, journal = {Applied Soft Computing}, month = {12}, pages = {113882}, shortjournal = {Appl. Soft. Comput.}, title = {OdML: An optimization-driven meta-learning framework for multi-task approximate model predictive control}, volume = {185}, year = {2025}, }
• (2025). A deep reinforcement learning trader without offline training. ASOC, 185, 113881. (www)

  @article{ , abstract = {In this paper we pursue the question of a fully online trading algorithm (i.e. one that does not need offline training on previously gathered data). For this task we consider Double Deep Q -learning in the episodic setting with Fast Learning Networks approximating the expected reward Q . Additionally, we define the possible terminal states of an episode in such a way as to introduce a mechanism to conserve some of the money in the trading pool when market conditions are seen as unfavourable. Some of these money are taken as profit and some are reused at a later time according to certain criteria. After describing the algorithm, we test it using 1-minute-tick price data for 4 major cryptocurrencies from Binance. We see that the agent performs better than trading with randomly chosen actions on each timestep. And it does so when tested on the whole dataset for a given market as well as on different subsets, representing different market trends.}, archive = {J_ASOC}, author = {Boian Lazov}, doi = {10.1016/j.asoc.2025.113881}, journal = {Applied Soft Computing}, month = {12}, pages = {113881}, shortjournal = {Appl. Soft. Comput.}, title = {A deep reinforcement learning trader without offline training}, volume = {185}, year = {2025}, }
• (2025). Density peak clustering algorithm based on boundary elimination and backbone construction. ASOC, 185, 113880. (www)

  @article{ , abstract = {Density peak clustering (DPC) is an effective clustering algorithm, but it still has some problems and faces some challenges. For instance, it cannot identify the variable density datasets, the assignment strategy is easy to produce domino phenomenon, and the clustering results of DPC and its improved algorithms are easily affected by the intersection points between clusters. To solve these problems, in this paper, we propose a novel density peak clustering algorithm based on boundary elimination and backbone construction, called BEBC-DPC. A new local density is defined based on the natural neighbor search, and the boundary degree is defined by the position relationship between each point and its neighbors, which accurately describes the local distribution information of the point. The boundary points of clusters are eliminated by fusing the density and the boundary degree, which reduces the influence of the intersection points on the cluster division. In addition, the cluster backbone construction method based on representative points and representative sets is proposed. The density relationship among non-boundary points is used to form representative sets, and the similarity between representative sets is used to construct the cluster backbones, which can effectively describe the overall distribution structure characteristics of the clusters. Moreover, the adjacency degree of each boundary point is defined by using the neighbor information and distance information, and the boundary points are gradually assigned to the most appropriate cluster backbone based on it to complete the clustering. Finally, sufficient experiments are performed on synthetic, UCI and image datasets, and the proposed BEBC-DPC is compared with DPC and its improved algorithms. Experimental results show the effectiveness of the proposed BEBC-DPC on various types of datasets.}, archive = {J_ASOC}, author = {Zhizhong Zhao and Sugen Chen and Cong Hu}, doi = {10.1016/j.asoc.2025.113880}, journal = {Applied Soft Computing}, month = {12}, pages = {113880}, shortjournal = {Appl. Soft. Comput.}, title = {Density peak clustering algorithm based on boundary elimination and backbone construction}, volume = {185}, year = {2025}, }
• (2025). Multi-criteria decision analysis-based framework for supply chain management evaluation with multi-dimensional sensitivity analysis: A green logistics perspective. ASOC, 185, 113879. (www)

  @article{ , abstract = {The transition toward sustainable Supply Chain Management (SCM) presents complex decision-making challenges that require robust and transparent evaluation methods. Effective decision-making in this context requires robust evaluation methods capable of handling complex, multi-dimensional aspects of the decision process. This study proposes a comprehensive decision-support framework that integrates Multi-Criteria Decision Analysis (MCDA) with advanced sensitivity analysis techniques to assess the stability and reliability of alternative rankings under uncertainty. The framework combines established MCDA methods with submodel exclusion analysis and the Comprehensive Sensitivity Analysis Method (COMSAM), which allows for systematic perturbation of the decision matrix and evaluation of robustness across multiple dimensions. Applied to a green logistics case study, the framework demonstrates its capacity to support sustainable decision-making by accounting for varying stakeholder priorities and data uncertainty. The approach offers methodological advances in robustness assessment, practical relevance for sustainability-focused SCM, and open-source implementation to ensure reproducibility. The proposed framework is adaptable to a wide range of decision contexts, contributing to the development of more reliable and explainable MCDA-based evaluations.}, archive = {J_ASOC}, author = {Jakub Więckowski and Wojciech Sałabun}, doi = {10.1016/j.asoc.2025.113879}, journal = {Applied Soft Computing}, month = {12}, pages = {113879}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-criteria decision analysis-based framework for supply chain management evaluation with multi-dimensional sensitivity analysis: A green logistics perspective}, volume = {185}, year = {2025}, }
• (2025). Reliable type 2 fuzzy Min–Max neural networks for pattern classification. ASOC, 185, 113875. (www)

  @article{ , abstract = {The Fuzzy Min–Max (FMM) algorithm is a powerful classification method capable of handling non-linear class boundaries and making both hard and soft decisions while learning from online data. However, it faces significant challenges, including sensitivity to the expansion coefficient, information loss during the contraction stage, and the overlap problem. To address these limitations, we propose a Reliable Type-2 Fuzzy Min–Max (RT2FMM) algorithm, which incorporates type-2 fuzzy logic to consider hyperbox uncertainty and effectively resolve the overlap problem. By assigning distinct certainties to overlapping regions, RT2FMM eliminates the need for the contraction stage and the overlap test. Additionally, we introduce weighted factors for hyperboxes, which enhances the reliability of membership values and models their mutual effects. Our comprehensive experimental evaluation across twenty datasets demonstrates that RT2FMM significantly outperforms existing FMM-based models in terms of robustness and accuracy. The Friedman test further confirms the superior performance of RT2FMM compared to commonly used classifiers, highlighting its potential as a robust solution for complex classification tasks.}, archive = {J_ASOC}, author = {Ali Nik-Khorasani and Mohammad-R. Akbarzadeh-T.}, doi = {10.1016/j.asoc.2025.113875}, journal = {Applied Soft Computing}, month = {12}, pages = {113875}, shortjournal = {Appl. Soft. Comput.}, title = {Reliable type 2 fuzzy Min–Max neural networks for pattern classification}, volume = {185}, year = {2025}, }
• (2025). Fortifying vision models: A comprehensive survey of defences against adversarial examples. ASOC, 185, 113874. (www)

  @article{ , abstract = {Artificial Intelligence and Machine learning (ML) have seen many advancements in the past two decades. It has led to the creation of several techniques, including Deep Neural Networks (DNN), Convolution Neural Networks (CNN), Autoencoders, Generative Adversarial Networks (GAN) and Diffusion models. These techniques have been applied to various real-world applications, such as self-driving cars, medical diagnosis and voice assistants. Despite these advancements, a carefully crafted input can fool the ML model. Such attacks are known as adversarial examples. It is a serious threat to safety critical systems. This survey provides a comprehensive review of defences against adversarial examples by tracing their evolution from early empirical methods to more principled, theoretically grounded approaches. We systematically categorise defences based on their underlying mechanisms. In addition to surveying state-of-the-art techniques, we spotlight emerging trends such as generative defences and diffusion-based purification. Finally, we identify persistent vulnerabilities and outline promising directions for future research towards building truly resilient vision models. This work aims to equip researchers and practitioners with a deep understanding of current defences and inspire innovation in adversarial robustness for the next generation of vision applications.}, archive = {J_ASOC}, author = {Siddheshwar Kumar and Shashank Srivastava and Shashwati Banerjea}, doi = {10.1016/j.asoc.2025.113874}, journal = {Applied Soft Computing}, month = {12}, pages = {113874}, shortjournal = {Appl. Soft. Comput.}, title = {Fortifying vision models: A comprehensive survey of defences against adversarial examples}, volume = {185}, year = {2025}, }
• (2025). AOT-PixelNet: Lightweight and interpretable detection of forged images via adaptive orthogonal transform. ASOC, 185, 113873. (www)

  @article{ , abstract = {Generative image detection faces persistent challenges in terms of generalization and interpretability, limiting its reliability in complex scenarios. To address these issues, we propose AOT-PixelNet, a lightweight and interpretable detection framework that integrates an Adaptive Orthogonal Transform (AOT) module with a streamlined 1 × 1 convolution-based PixelNet architecture. The AOT module leverages diverse orthogonal transforms, such as FFT and DCT, to extract informative frequency-domain features, thereby enhancing sensitivity to medium- and high-frequency artifacts. Meanwhile, PixelNet minimizes parameter count (only 0.98 million) while effectively capturing cross-channel inconsistencies and mitigating overfitting. Experimental evaluations on multiple unseen GAN and diffusion-based datasets demonstrate that AOT-PixelNet achieves superior performance with minimal computational cost. Specifically, it outperforms the NPR method by 0.6% and 11.76% on the ForenSynths and GenImage datasets, respectively, validating the framework’s robustness, effectiveness, and interpretability.}, archive = {J_ASOC}, author = {Dengtai Tan and Deyi Yang and Boao Tan and Chengyu Niu and Yang Yang and Shichao Li}, doi = {10.1016/j.asoc.2025.113873}, journal = {Applied Soft Computing}, month = {12}, pages = {113873}, shortjournal = {Appl. Soft. Comput.}, title = {AOT-PixelNet: Lightweight and interpretable detection of forged images via adaptive orthogonal transform}, volume = {185}, year = {2025}, }
• (2025). Status prediction using attentive and cascaded deep network by fault tolerant and priority-based scheduling for load balancing in cloud. ASOC, 185, 113872. (www)

  @article{ , abstract = {The cloud facilitates the user to complete their work utilizing the cost strategy of pay-as-you-go, which is based on the consumed Virtual Machine (VM) hours. Thus, the scheduler must offer the highest throughput to attain efficient allocation of resources in the cloud paradigm. Cloud services are dependent on characteristics such as fault tolerance, security, scalability, and availability. Hence, an effective scheduler is necessary to arrange the scheduling tasks and adjust the server loads. Typically, a load-balancing task focuses on detecting the overloaded and under-loaded nodes and adjusting the load between them. When considering the significant role of fault-tolerance in load-balancing algorithms, it seems to suffer from poor organization and a lack of in-depth experiments in this sector. This paper proposes a new task for the load-balancing operation. Initially, task scheduling is performed where the fault tolerance and the priority-aided scheduling approach are adopted. Furthermore, resource optimization is carried out in the scheduling task using Randomly Improved Electric Fish Optimization (RIEFO). To validate the load balancing operation, several multi-objective functions such as resource utilization, delay, time, makespan, active servers, throughput, success rate, fault tolerance rate, and energy consumption are derived. Moreover, because of the system’s dynamic environment, the status of the server varies simultaneously. The server status prediction is significant in allocating the tasks to the server or the VM resources. Thus, the Attention-based Cascaded Residual Bidirectional Long Short-Term Memory (ACRes-BiLSTM) is employed to predict the server status before performing the resource allocation. Finally, the tasks are scheduled effectively using the predicted server status. The performance is estimated using numerous performance metrics.}, archive = {J_ASOC}, author = {Gudivada Lokesh and Kasarapu Ramani and K.K. Baseer}, doi = {10.1016/j.asoc.2025.113872}, journal = {Applied Soft Computing}, month = {12}, pages = {113872}, shortjournal = {Appl. Soft. Comput.}, title = {Status prediction using attentive and cascaded deep network by fault tolerant and priority-based scheduling for load balancing in cloud}, volume = {185}, year = {2025}, }
• (2025). A hybrid deep learning framework for detecting grape leaf diseases by integrating ResNeSt-50 and CBAM attention into DeepLabv3 +. ASOC, 185, 113871. (www)

  @article{ , abstract = {Grape leaf diseases, such as black rot, powdery mildew, and downy mildew, pose a significant threat to global viticulture, leading to substantial yield losses and reduced fruit quality. Early and accurate identification of these diseases is essential for precision agriculture and sustainable crop management. This study presents a comprehensive comparison of traditional and deep learning-based image segmentation methods for detecting grape leaf lesions. A series of classical segmentation techniques, including Mean Shift, Fuzzy C-Means (FCM), Normalized Cut, K-Means, and Fuzzy K-Means (FKM), were evaluated alongside an advanced DeepLabv3 + model. The baseline DeepLabv3 + architecture was further enhanced by integrating a ResNeSt-50 backbone with various attention mechanisms, including Squeeze-and-Excitation (SE) Block, Convolutional Block Attention Module (CBAM), Bottleneck Attention Module (BAM), Self-Attention, and Dual Attention Network (DANet). Among all models, DeepLabv3 + with ResNeSt-50 and CBAM achieved the highest performance, attaining 98.2 % accuracy, 97.1 % precision, 96.7 % recall, 96.6 % mean Intersection over Union (mIoU), and a 96.8 % Dice Score. The results demonstrate that attention-augmented deep networks significantly outperform classical methods, especially in handling complex lesion structures under diverse environmental conditions. While traditional algorithms remain useful in resource-constrained scenarios, deep learning models, particularly those enhanced with spatial and channel-wise attention, offer greater accuracy and robustness, making them ideal for integration into intelligent agricultural platforms such as drones, mobile scanners, and automated disease monitoring systems. Future work will focus on incorporating temporal and multimodal data, expanding dataset diversity, and optimizing lightweight models for real-time deployment on edge devices.}, archive = {J_ASOC}, author = {Kittipol Wisaeng}, doi = {10.1016/j.asoc.2025.113871}, journal = {Applied Soft Computing}, month = {12}, pages = {113871}, shortjournal = {Appl. Soft. Comput.}, title = {A hybrid deep learning framework for detecting grape leaf diseases by integrating ResNeSt-50 and CBAM attention into DeepLabv3 +}, volume = {185}, year = {2025}, }
• (2025). A stochastic proportional–differential TLBO algorithm and its applications for wafer transfer finger. ASOC, 185, 113870. (www)

  @article{ , abstract = {The Teaching-Learning-Based optimization (TLBO) algorithm, which includes teacher phase and learner phase, is a widely used method for global optimization. However, TLBO will experience premature convergence and get stuck in local optimum when faced with complex optimization challenges. Especially when tackling complex problems in practical engineering applications, which involves multiple variables and numerous constraints. To address this issue, a new variant termed Stochastic Proportional–Differential TLBO (SPD-TLBO) has been developed. The SPD phase allows students to learn not only from the current population but also from previous stochastic errors and their generation differences using adaptive random operators. By incorporating an SPD operator into the original TLBO framework, the algorithm’s search diversity is enhanced, reducing the likelihood of premature convergence to local optimum. The experimental results conducted at the IEEE Conference on Evolutionary Computation 2014 (CEC 2014) indicated that the proposed SPD-TLBO algorithm achieved an effective balance between exploration and exploitation capabilities. Specifically, the SPD-TLBO algorithm achieves the highest ranking in 21 out of 30 cases (70%) for 30-dimensional problems and 18 out of 30 cases (60%) for 50-dimensional problems. Statistical tests and convergence analyses show that the SPD-TLBO algorithm outperforms other algorithms in solving global optimization problems. Additionally, when applied to engineering optimization problems, the SPD-TLBO algorithm shows significant advantages over other algorithms. Therefore, the SPD-TLBO algorithm is further applied to optimize the structure of a wafer transfer finger in semiconductor manufacturing.}, archive = {J_ASOC}, author = {Jinfeng Sun and Yunlang Xu and Haibo Zhou}, doi = {10.1016/j.asoc.2025.113870}, journal = {Applied Soft Computing}, month = {12}, pages = {113870}, shortjournal = {Appl. Soft. Comput.}, title = {A stochastic proportional–differential TLBO algorithm and its applications for wafer transfer finger}, volume = {185}, year = {2025}, }
• (2025). Prediction of the health effects of COVID using hybrid classifier with attention-based LSTM-deep temporal convolution network. ASOC, 185, 113867. (www)

  @article{ , abstract = {Coronavirus Disease 2019 (COVID-19) is an infectious illness that affects both humans and animals. Individuals infected with COVID-19 are prone to lung complications during the recovery phase . Radiography and Computed Tomography (CT) are the most commonly used methods for diagnosing lung-related diseases. The primary aim of this paper is to assess the impact of COVID-19 on patients’ lungs, heart, and blood sugar levels using a deep learning-based approach. Initially, data related to the heart, blood sugar levels, and lungs of COVID-19-infected individuals are collected. From this dataset, three types of features are extracted. Deep features are obtained using Iterated Dilated Convolutional Neural Networks (IDCNN). From these deep features, which are obtained from the IDCNN, the optimal weighted features are derived by implementing the Hybrid Dolphin Pod Cuttlefish Optimization (HDPCO) algorithm. Subsequently, the HDPCO algorithm is also employed for optimal feature extraction. In addition, dimensionality reduction is performed using Principal Component Analysis (PCA). These three sets of features from the IDCNN, HDPCO, and PCA, are then fused into a single feature set . This fused feature set is fed into a hybrid classifier composed of a Deep Temporal Convolutional Network (DTCN) and an Attention-based Long Short-Term Memory (ALSTM) network . The classifier parameters are optimized using the HDPCO algorithm. The output from the hybrid classifier provides the final prediction result. Experimental results demonstrate that the proposed COVID-19 impact prediction model significantly outperforms existing models in terms of prediction accuracy and efficiency.}, archive = {J_ASOC}, author = {Sadanandam Kalvala and B. Baranidharan}, doi = {10.1016/j.asoc.2025.113867}, journal = {Applied Soft Computing}, month = {12}, pages = {113867}, shortjournal = {Appl. Soft. Comput.}, title = {Prediction of the health effects of COVID using hybrid classifier with attention-based LSTM-deep temporal convolution network}, volume = {185}, year = {2025}, }
• (2025). Quantum fuzzy logic for edge detection: A demonstration on NISQ hardware. ASOC, 185, 113866. (www)

  @article{ , abstract = {Quantum computing offers the potential to enhance computational efficiency beyond classical methods, but practical implementation remains challenging due to the limitations of Noisy Intermediate-Scale Quantum (NISQ) hardware, namely, restricted qubit counts, limited connectivity, and the presence of noise and decoherence. This study presents a novel approach to edge detection by leveraging a recently developed Quantum Fuzzy Inference Engine, implemented on a NISQ device. We introduce an optimized quantum circuit for its implementation, reducing qubit requirements and gate depth to improve execution on NISQ hardware. To overcome constraints related to large-scale image processing, a hybrid quantum–classical lookup table approach is employed. Edge detection performance is evaluated on the Berkeley Segmentation Data Set and Benchmarks 500 dataset under different conditions, including classical execution, ideal quantum simulation, noisy quantum simulation, and NISQ hardware calculation. Results demonstrate that the quantum fuzzy logic-based edge detection achieves outcomes comparable to classical methods by using fewer operations, marking a step toward practical quantum-enhanced image processing.}, archive = {J_ASOC}, author = {G. Nunziata and S. Crisci and G. De Gregorio and R. Schiattarella and G. Acampora and L. Coraggio and N. Itaco}, doi = {10.1016/j.asoc.2025.113866}, journal = {Applied Soft Computing}, month = {12}, pages = {113866}, shortjournal = {Appl. Soft. Comput.}, title = {Quantum fuzzy logic for edge detection: A demonstration on NISQ hardware}, volume = {185}, year = {2025}, }
• (2025). Multi-agent cooperation-based bi-criteria evolutionary many-objective optimization. ASOC, 185, 113865. (www)

  @article{ , abstract = {Many-objective evolutionary algorithms (MaOEAs) excel in solving many-objective optimization problems (MaOPs), which are mainly classified into two frameworks: the Pareto domination and the non-Pareto domination. The Pareto criterion (PC) obtains a well-converged solution set in multi-objective spaces through the Pareto dominance relationship between solutions. However, insufficient environmental selection pressure in many-objective spaces leads to slow convergence. The non-Pareto criterion (NPC) enhances the selection pressure by evaluating the solution set with a set of sortable scalar values. However, it is difficult to ensure the Pareto-optimal consistency of convergence and distribution when facing highly irregular Pareto fronts (PFs). Therefore, combining the two sets of criteria can satisfy the demand for uniform distribution while bringing significant selection pressure. A multi-agent cooperative strategy is proposed in this study to realize the combination of the two criteria. This strategy controls the evolutionary direction of two populations separately by deploying two agents, and promotes cooperative evolution between these populations through the exchange and flow of large amounts of information. In order to better realize the cooperative effect, we adopt the multi-agent reinforcement learning (MARL) strategy to accurately regulate the variation operator and parameter configurations of the bi-population. In addition, the effectiveness of the proposed method is validated on 74 test problems (DTLZ, WFG, and UF) and 3 real-world problems. The results show that the proposed algorithm is more competitive than 6 state-of-the-art algorithms.}, archive = {J_ASOC}, author = {Jiazheng Li and Yuan Liu and Juan Zou and Shuyi Liu and Shengxiang Yang and Jinhua Zheng}, doi = {10.1016/j.asoc.2025.113865}, journal = {Applied Soft Computing}, month = {12}, pages = {113865}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-agent cooperation-based bi-criteria evolutionary many-objective optimization}, volume = {185}, year = {2025}, }
• (2025). Prompted complex context generation guided fine-grained ship recognition. ASOC, 185, 113856. (www)

  @article{ , abstract = {Fine-grained ship recognition in complex marine environments is challenged by background interference, high inter-class similarity, and limited labeled data. Existing methods often rely on inefficient cascades or holistic feature extraction, which limits both accuracy and efficiency. To address these issues, we propose a Prompted Complex Context Generation Guided Fine-Grained Ship Recognition framework, consisting of two core modules. The Cross-Attention Context Generation Module utilizes a diffusion model to generate diverse background images from prompts, maintaining target consistency and enriching the training data to mitigate data scarcity. It also employs a cross-attention map to highlight target-relevant regions, guiding the Attention Map Guided Fusion Module. The Attention Map Guided Fusion Module adopts a dual-branch transformer architecture: one branch extracts global features from background-enhanced images, and the other captures local features through attention-guided cropping of target-specific regions. By integrating both global and local features, our method effectively identifies key target characteristics. Experimental results demonstrate that our approach achieves 97.04% accuracy on the publicly available MAR-ships dataset and 84.57% accuracy on the challenging GCS dataset, outperforming state-of-the-art methods.}, archive = {J_ASOC}, author = {Runtian Wang and Kejun Wu and Renjie Qiao and Chunsheng Yang and Chengtao Cai}, doi = {10.1016/j.asoc.2025.113856}, journal = {Applied Soft Computing}, month = {12}, pages = {113856}, shortjournal = {Appl. Soft. Comput.}, title = {Prompted complex context generation guided fine-grained ship recognition}, volume = {185}, year = {2025}, }
• (2025). Multi-label feature selection via asymmetric redundancy and variable precision dependency. ASOC, 185, 113852. (www)

  @article{ , abstract = {Multi-label feature selection is an effective data preprocessing technique that can significantly mitigate the challenges posed by high-dimensional features in multi-label learning. However, the exploration of feature-label correlations has often been strictly limited to inclusion relationships, while ignoring the fusion of local and global label information. Moreover, most previous work has typically assumed that redundancy between features is fully symmetric, overlooking the valuable insights that asymmetric redundancy provides for designing feature selection. To address these issues, this paper proposes a novel multi-label feature selection via asymmetric redundancy and variable precision dependency. Specifically, it constructs a conditional probability model to reflect the local label semantics, incorporating this into the construction of the variable precision dependency through a fusion indicator. Subsequently, the optimistic and pessimistic information overlap between features is discussed, allowing variable precision granularity to capture asymmetric redundancy between features. Building upon this, an information fusion method is proposed to quantify the pessimistic asymmetric redundancy between features by inducing knowledge granularity in the feature space. Finally, a comprehensive evaluation metric, Maximum Correlation-maximum Discrimination-minimum Redundancy (MCDR), is proposed to evaluate the significance of features. The experimental results on fifteen multi-label benchmark datasets indicate that the proposed method outperforms the other seven state-of-the-art methods.}, archive = {J_ASOC}, author = {Wenbin Qian and Xiwen Lu and Shiming Dai and Jintao Huang}, doi = {10.1016/j.asoc.2025.113852}, journal = {Applied Soft Computing}, month = {12}, pages = {113852}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-label feature selection via asymmetric redundancy and variable precision dependency}, volume = {185}, year = {2025}, }
• (2025). A multi-granularity decision tree algorithm based on variable precision rough sets and zentropy. ASOC, 185, 113851. (www)

  @article{ , abstract = {The existing decision tree algorithms often use a single-layer measure to process data, which cannot fully consider the complex interactions and dependencies between different granularity levels. In addition, decision tree algorithms inevitably face the issue of multi-value preference, which may lead to the selection of unreasonable attributes in the process of partition, thereby affecting the performance of the algorithms. Therefore, this paper proposes an improved decision tree algorithm, called Ze-VNDT, which combines variable precision rough sets with Zentropy. First, to avoid the information loss caused by data discretization, this paper introduces variable precision neighborhood rough sets for data processing. Second, by analyzing the granularity level structure within the variable precision neighborhood rough set model, knowledge uncertainty is analyzed from three granularity levels: decision classes, approximate relations, and similarity classes. We describe the uncertain knowledge from the overall to the internal using the idea of going from coarse to fine, and design a Zentropy to measure uncertainty. To address the issue of multi-value preference, an adaptive weighted Zentropy uncertainty measure is designed based on the definition of uncertainty measure based on Zentropy. Third, when constructing the improved decision tree algorithm, the optimal attributes are selected based on the designed uncertainty measure. Finally, numerical experiments on 18 UCI datasets validated the effectiveness and rationality of the proposed algorithm. The experimental results showed that, compared to traditional algorithms and the latest improved algorithms, the proposed algorithm achieved an average accuracy of 94.79%, an average precision of 85.77%, an average recall rate of 84.68%, and an F1-score of 84.97% across the 18 datasets. It ranked first in all five evaluation metrics, demonstrating higher stability and accuracy.}, archive = {J_ASOC}, author = {Hui Dong and Caihui Liu and Xiying Chen and Duoqian Miao}, doi = {10.1016/j.asoc.2025.113851}, journal = {Applied Soft Computing}, month = {12}, pages = {113851}, shortjournal = {Appl. Soft. Comput.}, title = {A multi-granularity decision tree algorithm based on variable precision rough sets and zentropy}, volume = {185}, year = {2025}, }
• (2025). Skin lesion classification with mini-batch sampling and deep metric learning. ASOC, 185, 113850. (www)

  @article{ , abstract = {Skin lesion image classification based on deep learning has recently garnered significant attention. However, directly applying methods that perform well in general computer vision tasks to skin lesion image classification is not ideal, as skin lesion image datasets possess intrinsic characteristics, such as class imbalance, intra-class variability, and inter-class similarity. To tackle these challenges simultaneously, we propose a novel unified learning framework, named mBSML, which integrates mini-batch sampling and deep metric learning. In this framework, mini-batch sampling re-samples data in real-time during each iteration of learning, while a new loss function combines mini-batch distance metric-based loss with cross-entropy loss. Through the alternating training procedure on both imbalanced training data and balanced re-sampling data, mBSML effectively learns from global distribution information and local similarity information, not only from the original dataset but also from the minority classes. Extensive experiments conducted on two publicly available datasets demonstrate the effectiveness of mBSML for skin lesion image classification.}, archive = {J_ASOC}, author = {Shengdan Hu and Zhifei Zhang and Li Ying and Guangming Lang}, doi = {10.1016/j.asoc.2025.113850}, journal = {Applied Soft Computing}, month = {12}, pages = {113850}, shortjournal = {Appl. Soft. Comput.}, title = {Skin lesion classification with mini-batch sampling and deep metric learning}, volume = {185}, year = {2025}, }
• (2025). AOD-YOLO: A self-modulating multi-scale feature aggregation mechanism for small object detection in airport surface scenes. ASOC, 185, 113849. (www)

  @article{ , abstract = {Object detection in airport surface scenes is crucial for enhancing safety. However, the coexistence of objects with significant scale disparities within the same region complicates feature representation, limiting existing models’ ability to capture fine-grained details, especially for small objects. To address this challenge, we propose AOD-YOLO, an Airport Object Detection (AOD) model incorporating a Self-Modulating Multi-Scale Feature Aggregation Mechanism. This model introduces two key innovations: (1) Enhanced Context Modeling: By leveraging large-kernel convolution, frequency-domain modulation, and statistical feature analysis, our approach effectively adjusts feature contributions across different object scales, improving contextual understanding in complex scenes; (2) Optimized Small Object Representation: A dynamic gradient gain allocation strategy refines small-object features, enhancing detection accuracy and overall feature presentation. AOD-YOLO consistently improves performance across model scales. On our self-constructed Airport dataset and the public VisDrone-DET2019 dataset, it achieves mean Average Precision (mAP 0.5 ) of 87.9% and 44.9%, respectively—outperforming state-of-the-art models like YOLOv11 and Gold-YOLO by substantial margins. Additionally, through optimized network module placement, AOD-YOLO achieves 112 FPS, striking a balance between computational efficiency and accuracy, making it well-suited for real-time airport object detection.}, archive = {J_ASOC}, author = {Yingqing Wang and Weili Zeng and Ziyu Zhao and Baogeng Li and Zhibin Quan}, doi = {10.1016/j.asoc.2025.113849}, journal = {Applied Soft Computing}, month = {12}, pages = {113849}, shortjournal = {Appl. Soft. Comput.}, title = {AOD-YOLO: A self-modulating multi-scale feature aggregation mechanism for small object detection in airport surface scenes}, volume = {185}, year = {2025}, }
• (2025). Gloss-free sign language translation based on fusion attention. ASOC, 185, 113848. (www)

  @article{ , abstract = {Sign language translation converts sign language videos into spoken language text. Traditional gloss-based approaches require costly gloss annotations, driving recent interest in gloss-free methods. In this paper, we propose a novel gloss-free sign language translation method based on fusion attention (SLTFA) that uniquely models the intrinsic logical structure of sign language. The key innovation is the development of a dual-attention mechanism that mimics the natural hierarchical structure of sign language: intra-gloss attention captures fine-grained relationships within video frame sequences representing individual semantic units, while inter-gloss attention models the broader contextual connections between these units, similar to how words form coherent sentences. Additionally, we introduce a contrastive loss strategy for cross-modal soft alignment that effectively bridges the gap between visual and textual representations. Extensive experiments on the RWTH-PHOENIX-WEATHER-2014T dataset demonstrate SLTFA’s superior performance, achieving a BLEU-4 score of 16.99 and a ROUGE score of 40.82. On the CSL-Daily dataset, our approach achieves a BLEU-1 score of 25.56 and a ROUGE score of 27.51, demonstrating strong performance across different sign languages.}, archive = {J_ASOC}, author = {Yingchun Xie and Wei Su and Chongliang Zhong and Chuan Cai and Yongna Yuan}, doi = {10.1016/j.asoc.2025.113848}, journal = {Applied Soft Computing}, month = {12}, pages = {113848}, shortjournal = {Appl. Soft. Comput.}, title = {Gloss-free sign language translation based on fusion attention}, volume = {185}, year = {2025}, }
• (2025). An efficient mathematical-based optimization method to optimize multi-hydropower operating rules. ASOC, 185, 113846. (www)

  @article{ , abstract = {Optimizing hydropower multi-reservoir systems requires both effective operating rules and efficient optimization techniques. The main contribution of this paper is offering a unique approach that elegantly combines two important parts: creating an efficient optimization method and developing hydropower operating rules. In this regard, a nonlinear rule curve (NLRC) and a linear rule curve (LRC), are tailored for the coordination of a hydropower multi-reservoir system (HMRS) in Iran. To optimize operating rules, the study fabricates a novel algorithm termed the multi-operator weighted mean of vectors (MINFO). The algorithm combines a powerful global search strategy (GSS) that thoroughly searches the solution space with an efficient local search (LS), striking a balance between solution diversity and convergence speed. To fine-tune this balance, an adaptive parameter-tuning strategy is applied. Furthermore, the active-set sequential quadratic programming (ASQP) serves as a localized escaping operator to enhance the algorithm's convergence speed. The effectiveness of the proposed MINFO algorithm is first evaluated through a nonlinear five-reservoir problem. The findings indicate that the MINFO algorithm outperforms a set of 14 distinct optimization methods. Subsequently, the MINFO algorithm is applied to identify optimal NLRC and LRC for a six-reservoir hydropower system. The results underscore the superiority of optimized NLRC, yielding a potential power augmentation of up to 17 % in comparison to the LRC approach. In summation, this study constitutes a seminal contribution by cultivating an efficient rule curve framework for the management of HMRSs.}, archive = {J_ASOC}, author = {Shuguang Li and Iman Ahmadianfar and Aitazaz A. Farooque and Zaher Mundher Yaseen}, doi = {10.1016/j.asoc.2025.113846}, journal = {Applied Soft Computing}, month = {12}, pages = {113846}, shortjournal = {Appl. Soft. Comput.}, title = {An efficient mathematical-based optimization method to optimize multi-hydropower operating rules}, volume = {185}, year = {2025}, }
• (2025). Advanced YOLOv4 for real-time underwater object detection: An application-oriented approach. ASOC, 185, 113837. (www)

  @article{ , abstract = {Detecting debris and monitoring marine life in sea aquaculture face challenges due to limited visibility and the presence of diverse. Underwater object detection by Autonomous Unmanned Vehicle(AUV) is inherently more challenging than land due to light attenuation and water turbidity, especially for small and dense objects in murky images, where extracting high-quality features is hindered. In this paper, we present an efficient approach for real-time underwater object detection through improvements in image enhancement, data augmentation, and feature aggregation. Initially, U-Shape Transformer is applied to enhance the original images. For data augmentation, it is observable that while Mosaic data augmentation enhances complex images but fails to improve small-object detection due generation of less number of images with small objects. To address this limitation, we propose Underwater-Mosaic (U-Mosaic), a modified Mosaic data augmentation technique designed to enhance small-object detection. Additionally, it was noted that existing YOLOv4 struggles with detecting small and densely populated objects in underwater images as unable to get sufficient features for small objects due to downsampling, image quality and also found difficulty in selecting anchor box size. Therefore, we propose a model called Advanced YOLOv4, tailored for underwater object detection. The proposed Advanced YOLOv4 aims to improve object detection efficiency by altering the neck and prediction layers of YOLOv4. Moreover, we introduce an additional spatial pyramid pooling layer to aggregate features and reduce feature dimensions thereby improving object detection rates. Also, the proposed work concentrates on very large object detection and for this purpose used downsampling during the detection of large objects. The proposed approach is validated through two distinct application areas: (i) detecting and locating debris (ii) detecting fish from underwater images. For validation, the Trash ICRA19 dataset is used for debris detection, while the Brackish dataset is employed for fish detection. UIQM and UCIQE, image enhancement assessment metrics are used to measure quality of enhanced images and found more than 20% better result for both the datasets. The proposed real-time underwater object detection model outperformed single-stage object detectors like YOLOv3, YOLOv4, YOLOv5, YOLOv7, and KPE-YOLOv5 by 5% in terms of mean Average Precision(mAP). Also proposed work compared with two-stage detector RCNN and found 8% better mAP than RCNN.}, archive = {J_ASOC}, author = {Pratima Sarkar and Sourav De and Prasenjit Dey and Sandeep Gurung}, doi = {10.1016/j.asoc.2025.113837}, journal = {Applied Soft Computing}, month = {12}, pages = {113837}, shortjournal = {Appl. Soft. Comput.}, title = {Advanced YOLOv4 for real-time underwater object detection: An application-oriented approach}, volume = {185}, year = {2025}, }
• (2025). UniCon: Unified image-guiding generation with noise consistency. ASOC, 185, 113832. (www)

  @article{ , abstract = {Diffusion models have demonstrated remarkable capabilities in image-to-image tasks. However, existing methods typically focus either on structural (e.g., layout, content) or stylistic guidance, with few approaches effectively excel at both. On the other hand, many methods require time-consuming fine-tuning or high inference latency, making interactive generation applications challenging to realize. To address these issues, we propose a two-stage framework referred as UniCon ( Uni fied Image-guiding Generation with Noise Con sistency). To improve time efficiency, we follow the paradigm of inversion-based image manipulation and introduce a novel method called Noise Consistency Inversion . Leveraging the nature of Consistency Models, this inversion process is highly efficient, requiring only a single neural function evaluation (NFE) in the inversion process. To achieve high consistency and finer control, we introduce a unified attention-based guidance mechanism that supports structural, stylistic, or joint reference inputs, without any additional fine-tuning. Experiments with structure- and style-specific methods show that our approach performs competitively or better in each individual aspect. In comparison of style transfer tasks that demand both structure and style, our method outperforms state-of-the-art baselines, confirming the effectiveness of our union control strategy. And overall, our approach also achieves the best efficiency in terms of runtime performance.}, archive = {J_ASOC}, author = {Yuanjun Liao and Yuning Gong and Yanci Zhang}, doi = {10.1016/j.asoc.2025.113832}, journal = {Applied Soft Computing}, month = {12}, pages = {113832}, shortjournal = {Appl. Soft. Comput.}, title = {UniCon: Unified image-guiding generation with noise consistency}, volume = {185}, year = {2025}, }
• (2025). Evaluation of the financial performance of the textile and apparel industry in interval type-2 fuzzy environment. ASOC, 185, 113830. (www)

  @article{ , abstract = {The Turkish textile and apparel sector plays a crucial role in the national economy through employment, exports, and investment. The financial performance of companies is a key determinant of their sustainability and competitiveness, especially in global markets. The Turkish textile and apparel sector is one of the essential industries in terms of macro-economic indicators such as net foreign exchange inflow, employment and investment. This sector is also one of the critical actors in world trade. A robust performance evaluation model is essential for stakeholders such as investors, creditors, and managers. However, the assessment of firms is a very critical decision involving uncertainty due to various conflicting criteria based on judgements. In this study, an integrated multi-criteria decision-making (MCDM) model including interval type-2 fuzzy hierarchy process (IT2FAHP) and Compromise Ranking of Alternatives from Distance to Ideal Solution (CRADIS) approaches are proposed to assess the financial performance of Turkish textile and clothing firms that are traded in Borsa İstanbul (BİST) in the period from 2006 to 2020. In line with the determined purpose, the arithmetic average of the determined financial ratios during the analysis period covering 15 years is computed to obtain long-term performance indicators. The importance weights of the selected financial criteria for the performance evaluation model are identified by employing the IT2FAHP approach. Then, the firms are ranked according to their financial performances with the CRADIS method. In addition, the results from the sensitivity analysis validate the proposed approach and prove that it is practical. Moreover, practical and managerial implications are discussed based on the results. The results offer valuable insights for strategic decision-making and can support efforts to enhance financial stability in the textile and apparel sector. According to the results, "LUKSK" had the highest long-term financial performance among the 11 companies discussed. This company is followed by BOSSA, YATAS, and ATEKS companies. The alternatives confirm the robustness of the proposed model in maintaining its place in the ranking in 190 scenarios. In addition, the comparative analysis confirms the consistency of the proposed ranking framework.}, archive = {J_ASOC}, author = {Ömer Faruk Görçün and Mohsin Shabir and Ahmet Çalık and Özcan Işık}, doi = {10.1016/j.asoc.2025.113830}, journal = {Applied Soft Computing}, month = {12}, pages = {113830}, shortjournal = {Appl. Soft. Comput.}, title = {Evaluation of the financial performance of the textile and apparel industry in interval type-2 fuzzy environment}, volume = {185}, year = {2025}, }
• (2025). An interpretable dual-output multivariate wind speed interval prediction scheme using hyper-heuristic optimization algorithm and deep neural networks. ASOC, 185, 113829. (www)

  @article{ , abstract = {Uncertainty analysis of wind speed forecasting using the Lower Upper Bound Estimation (LUBE) represents an advanced interval prediction method that does not require assumptions about data distribution. Previous studies, however, have exclusively focused on univariate prediction models, neglecting the information from other variables, and have not fully exploited the prediction errors in their loss function during training. To address these issues, an interpretable dual-output multivariate wind speed interval prediction scheme (IMWSIPS) that utilizes a hyper-heuristic optimization algorithm and a deep neural network is proposed, along with a novel loss function for training. The system initially takes multiple inputs such as historical wind speed and other influencing factors including wind direction, density, temperature, and pressure into a deep neural network. The actual wind speeds are then scaled up and down by factors of 1 + θ 1 (0 <θ 1 <1) and 1 + θ 2 (-1 <θ 2 <0), respectively, to produce two outputs from the network. On this basis, an optimization problem to minimize interval width under a given coverage probability is formulated and solved using the developed hyper-heuristic algorithm, yielding optimal values for θ 1 and θ 2 and the prediction intervals for sub-models. Subsequently, the advantages of five deep neural network models are leveraged to construct an ensemble model, with weights optimized by the hyper-heuristic algorithm to derive the final prediction intervals. Ultimately, the system's interpretability is analyzed at both variable and sub-model levels. Experimental and discussion results demonstrate that the introduction of IMWSIPS not only signifies enhancements in forecasting performance but also implies improvements in wind energy utilization efficiency and reductions in operational costs for power systems.}, archive = {J_ASOC}, author = {Mengzheng Lv and Jianzhou Wang and Shuai Wang and Yang Zhao and Jialu Gao and Yuansheng Qian}, doi = {10.1016/j.asoc.2025.113829}, journal = {Applied Soft Computing}, month = {12}, pages = {113829}, shortjournal = {Appl. Soft. Comput.}, title = {An interpretable dual-output multivariate wind speed interval prediction scheme using hyper-heuristic optimization algorithm and deep neural networks}, volume = {185}, year = {2025}, }
• (2025). The fuzzified grey wolf: An improved grey wolf optimizer based on dynamic fuzzy system FGWO. ASOC, 185, 113818. (www)

  @article{ , abstract = {The Grey Wolf Optimizer (GWO) is a leading, powerful, and effective algorithm in swarm optimization techniques, showing competitive performance across various optimization problems. Yet, GWO is burdened by high tendencies toward exploitation and imprecise population diversity. This work introduces an improved GWO called the Fuzzified Grey Wolf Optimizer (FGWO) for solving global optimization problems. FGWO benefits from a dynamic fuzzy inference system (DFIS) to capture the optimal value of a → throughout iterations. DFIS integrates two inputs: the diversity rate and iteration number, and by inferring the optimal value of a → , FGOW determines whether to exploit or explore. Moreover, DFIS employs an adaptive membership function to capture the precise value of population diversity throughout the course of iteration. This optimal a → determination strategy can achieve a balanced exploitation–exploration ratio, mitigating premature convergence and enhancing diversity. FGWO is evaluated on CEC2017 benchmark functions, four engineering designs, and a breast cancer genes feature selection design. FGWO was compared with three other improved GWOs, and across all experiments, the outcomes demonstrate its superiority in terms of efficiency and applicability to real-world designed problems.}, archive = {J_ASOC}, author = {Mohammed Dheyaa Algubili and Labiba M. Alhelfi and Hana’ M. Ali}, doi = {10.1016/j.asoc.2025.113818}, journal = {Applied Soft Computing}, month = {12}, pages = {113818}, shortjournal = {Appl. Soft. Comput.}, title = {The fuzzified grey wolf: An improved grey wolf optimizer based on dynamic fuzzy system FGWO}, volume = {185}, year = {2025}, }
• (2025). Fuzzy lagrange interpolation method from summation of interactive fuzzy numbers. ASOC, 185, 113817. (www)

  @article{ , abstract = {This article proposes a novel approach to extending the sum of interactive fuzzy numbers, which is independent of the order of its operands. Interactive fuzzy numbers are fuzzy quantities in which the values across different α -cuts are not assumed to vary independently, incorporating dependencies that better reflect real-world uncertainty. A characterization of the proposed summation is given in terms of α -cuts, making computational implementation easier. It is shown that this operation preserves essential mathematical properties, including associativity. This is particularly important, as it enables the consistent aggregation of multiple fuzzy quantities without concern for the order in which the operands are grouped. The norm and width behaviors under this new summation are also analyzed. To illustrate the theoretical results, several examples are provided. As a practical application, the classical Lagrange polynomial interpolation method is extended to handle uncertain parameters represented by interactive fuzzy numbers. A fuzzy curve fitting problem is examined using this framework, and a comparative discussion highlights the advantages of the proposed method over existing approaches.}, archive = {J_ASOC}, author = {Geizane Lima da Silva and Estevão Esmi and Vinícius Francisco Wasques and Laécio Carvalho de Barros}, doi = {10.1016/j.asoc.2025.113817}, journal = {Applied Soft Computing}, month = {12}, pages = {113817}, shortjournal = {Appl. Soft. Comput.}, title = {Fuzzy lagrange interpolation method from summation of interactive fuzzy numbers}, volume = {185}, year = {2025}, }
• (2025). Bridging the semantic-numerical gap: A numerical reasoning method of cross-modal knowledge graph for material property prediction. ASOC, 185, 113776. (www)

  @article{ , abstract = {Material properties are illustrated by numerical data and semantic factors. In general, existing methods typically adopt machine learning (ML) algorithms to regress numerical properties or transfer other pre-trained knowledge graphs (KGs) to the material, due to the limitations of small-sample datasets. However, integrating semantic and numerical information from multi-modal data which across diverse experimental conditions remains a significant challenge in materials science. In this paper, a numerical reasoning method for material KGs (NR-KG) 1 was proposed, which constructs a cross-modal KG using semantic nodes and numerical proxy nodes. Both types of information by projecting KG into a canonical KG were captured and a graph neural network to predict material properties was utilized. In process, a novel projection prediction loss is proposed to extract semantic features from numerical information. NR-KG facilitates end-to-end processing of cross-modal data, mining relationships and cross-modal information in small-sample datasets, and fully utilizes effective experimental data to enhance the accuracy of material prediction. We propose two new high-entropy alloys (HEA) property datasets with semantic descriptions. NR-KG outperforms state-of-the-art (SOTA) methods on two material datasets, with MSE values of 3520 and 2.210, and achieving relative improvements of 25.9% and 16.1%, respectively, over the second-best methods, KANO and PCHMLP (semantic). It also achieves RMSE values of 0.584 and 0.521 on the FreeSolv and ESOL public molecular datasets, surpassing SOTA methods by 48.8% and 22.2% over KANO, highlighting its potential application and generalizability.}, archive = {J_ASOC}, author = {Guangxuan Song and Dongmei Fu and Zhongwei Qiu and Zijiang Yang and Jiaxin Dai and Lingwei Ma and Dawei Zhang}, doi = {10.1016/j.asoc.2025.113776}, journal = {Applied Soft Computing}, month = {12}, pages = {113776}, shortjournal = {Appl. Soft. Comput.}, title = {Bridging the semantic-numerical gap: A numerical reasoning method of cross-modal knowledge graph for material property prediction}, volume = {185}, year = {2025}, }
• (2025). Transformer-based actor–critic for adaptive cryptocurrency portfolio rebalancing. ASOC, 185, 113697. (www)

  @article{ , abstract = {High-frequency trading in volatile markets, such as cryptocurrencies, requires portfolio models that can swiftly adapt to regime shifts while controlling risk. We propose a novel approach that frames portfolio management as a dynamic strategy-selection problem. Instead of directly predicting asset weights, our agent selects from a pool of expert strategies based on recent market trends. We introduce a Transformer-based Variational Autoencoder (VAE) to extract disentangled trend representations, and a trend-aware actor–critic model to perform expert selection. Experiments demonstrate that this modular, strategy-level control mechanism outperforms existing methods in risk-sensitive crypto portfolio management.}, archive = {J_ASOC}, author = {Ahmad Asadi and Reza Safabakhsh}, doi = {10.1016/j.asoc.2025.113697}, journal = {Applied Soft Computing}, month = {12}, pages = {113697}, shortjournal = {Appl. Soft. Comput.}, title = {Transformer-based actor–critic for adaptive cryptocurrency portfolio rebalancing}, volume = {185}, year = {2025}, }
• (2025). Sustainable bridge management using refined feature selection for machine learning-aided bridge condition prediction: Incorporation of pareto distribution in MRMR method. ASOC, 184, 113878. (www)

  @article{ , abstract = {One of the critical phases of bridge management is condition prediction, which was enhanced after machine learning emerged. Researchers have used feature selection (FS) methods to reduce the predictors and optimise the prediction models. Prior studies have either explored a unified feature set for the overall bridge condition or focused solely on the deck, leading to limited predictions. This study proposes a refined feature selection method highlighting the importance of specific predictors for different bridge elements’ conditions using the California State inspection database from 1983 to 2021. The implemented FS approach consists of a verified Minimum-Redundancy Maximum-Relevance (MRMR) method followed by a Pareto analysis that identifies the most contributory factors in predicting the condition of bridges’ decks, superstructures, and substructures. The applied experiment on the US National Bridge Inventory database reveals that 28–33 predictor variables, out of more than 140 available features, contribute the most to each component’s health prediction with a cumulative importance score of over 95 %. Additionally, 22 mutual data items among the selected features are proposed as the minimum required predictors to be gathered by the asset management authorities. This study’s achievements help both researchers reduce the running costs of their prediction models and asset managers with data gathering and registration optimisation and, consequently, whole-of-life cycle cost reduction for sustainable asset management.}, archive = {J_ASOC}, author = {Vandad Dayan and Nicholas Chileshe and Reza Hassanli and Amin Parvaneh}, doi = {10.1016/j.asoc.2025.113878}, journal = {Applied Soft Computing}, month = {12}, pages = {113878}, shortjournal = {Appl. Soft. Comput.}, title = {Sustainable bridge management using refined feature selection for machine learning-aided bridge condition prediction: Incorporation of pareto distribution in MRMR method}, volume = {184}, year = {2025}, }
• (2025). Development of fuzzy rule-based models in the presence of the big data environment. ASOC, 184, 113869. (www)

  @article{ , abstract = {In this study, we propose a series of methods to build fuzzy rule-based models (FRBMs) in the presence of the big data environment such that the formed predictive models are more accurate, efficient, and robust. We follow two major steps to realize this target. In the first step, we build numeric FRBMs with the big data set such that the formed predictive models are more accurate and efficient. Specifically, based on the divide-and-conquer strategy, the big data set is divided into subsets through either the hyperplane division-based method or the K -Means clustering-based method; then either a global-based strategy or a local-based strategy is used to build numeric FRBMs. As a result, four Options are generated to develop numeric FRBMs. In the second step, we build the granular FRBMs based on the four Options developing the numeric FRBMs. Specifically, given a certain Option, based on the Principle of Justifiable Granularity (PJG), we granulate both condition parts and conclusion parts of the rules, forming the granular FRBMs; then the predictive models are further evaluated based on the PJG and optimized based on the Particle Swarm Optimization (PSO) algorithm to enhance the robustness. Finally, experimental studies on both synthetic datasets and publicly available datasets are conducted to prove the effectiveness of the proposed methods.}, archive = {J_ASOC}, author = {Yinghua Shen and Dan Zhao and Yan Li and Xingchen Hu and Yuan Chen and Bingsheng Liu}, doi = {10.1016/j.asoc.2025.113869}, journal = {Applied Soft Computing}, month = {12}, pages = {113869}, shortjournal = {Appl. Soft. Comput.}, title = {Development of fuzzy rule-based models in the presence of the big data environment}, volume = {184}, year = {2025}, }
• (2025). CMFF: A cross-modal multi-layer feature fusion network for multimodal sentiment analysis. ASOC, 184, 113868. (www)

  @article{ , abstract = {Multimodal sentiment analysis seeks to interpret speaker sentiment by integrating information from multiple modalities, typically text and audio. While existing methods often focus on fusing deep-layer features extracted from the final stages of unimodal encoders, they may overlook crucial fine-grained information present in shallow-layer features (e.g., subtle phonetic variations or basic syntactic structures) relevant for nuanced sentiment understanding. Furthermore, effectively fusing features from different modalities presents the dual challenges of dynamically weighting each modality’s contribution and accommodating their inherent data heterogeneity. To address these limitations, we propose a novel Cross-modal Multi-layer Feature Fusion (CMFF) network. CMFF explicitly leverages the hierarchical information contained in both shallow-layer and deep-layer features from text and audio modalities. It employs multi-head cross-modal attention mechanisms within its fusion layers to facilitate interaction across feature layers and modalities. Crucially, CMFF incorporates a Mixture of Gated Experts (MoGE) network within these fusion layers. The MoGE utilizes modality-specific expert sub-networks, each tailored to process the distinct characteristics of text or audio data, thereby directly addressing data heterogeneity. Concurrently, each expert employs an internal gated feed-forward mechanism. This allows the model to dynamically control the information flow for each feature vector, effectively learning to weigh the importance of different feature dimensions from each layer and modality based on the input context. Extensive experiments conducted on the benchmark CMU-MOSI and CMU-MOSEI datasets demonstrate that the proposed CMFF model achieves competitive or superior performance compared to state-of-the-art methods across various standard evaluation metrics.}, archive = {J_ASOC}, author = {Shuting Zheng and Jingling Zhang and Yuanzhao Deng and Lanxiang Chen}, doi = {10.1016/j.asoc.2025.113868}, journal = {Applied Soft Computing}, month = {12}, pages = {113868}, shortjournal = {Appl. Soft. Comput.}, title = {CMFF: A cross-modal multi-layer feature fusion network for multimodal sentiment analysis}, volume = {184}, year = {2025}, }
• (2025). Mitigating over-squashing in graph few-shot learning by leveraging local and global similarities. ASOC, 184, 113863. (www)

  @article{ , abstract = {Supervised machine learning models, particularly neural networks, often fail to deliver satisfactory results in scenarios with insufficient data. This becomes even more challenging when dealing with inherently complex data, such as graph data. This paper addresses the issue of learning with a limited number of samples, known as n -way k -shot learning, within the context of graph data. Our research extends the concept of similarity from neighboring nodes to the entire graph by leveraging transitivity relations. By employing edges and strong transitivity relations, we utilize a bipartite graph neural network that capitalizes on both local neighborhoods and distant, yet similar, nodes to generate node embeddings. This approach has demonstrated effectiveness in tasks such as node classification. Our proposed model’s ability to mitigate the over-squashing problem enhances its generalizability, resulting in a task-invariant model. Experimental results on various graph datasets show that the embeddings produced by our model are not task-specific. Consequently, our model outperforms other models in few-shot learning scenarios, where only a limited number of labeled nodes are available for each distinct downstream task.}, archive = {J_ASOC}, author = {Yassin Mohamadi and Mostafa Haghir Chehreghani}, doi = {10.1016/j.asoc.2025.113863}, journal = {Applied Soft Computing}, month = {12}, pages = {113863}, shortjournal = {Appl. Soft. Comput.}, title = {Mitigating over-squashing in graph few-shot learning by leveraging local and global similarities}, volume = {184}, year = {2025}, }
• (2025). Causal intervention for feedback information in fairness recommendation. ASOC, 184, 113862. (www)

  @article{ , abstract = {In recommender systems, users’ feedback information is greatly affected by individual differences between users and usually interferes with the interaction between users and items, causing unfairness in the recommendation results. To alleviate the problem of unfair recommendation results caused by the confounding of feedback information, we propose a novel causal intervention for feedback information in fairness recommendation (CIFair). First, we construct a causal graph based on the interaction between users and items and analyze the reasons for unfair recommendation results caused by feedback information as the confounding factor through the causal graph. Then, we design a two-phase predicted rating generation, namely, the elimination and reconstruction phases. In the elimination phase, we analyze the dual effects of the confounding factor on the recommender system and eliminate it through causal intervention to improve the fairness of the recommendation results and obtain a fair predicted rating. In the reconstruction phase, we design personalized feedback information based on user and item attributes to ensure recommendation performance and obtain a personalized predicted rating. Finally, we combine the predicted ratings generated in the elimination and reconstruction phases and provide users with personalized recommendation results with fairness. We conduct experiments on three publicly available datasets (MovieLens-1M, Last.fm, and Yelp) to verify the significance of the CIFair. The ablation experiment confirms that the CIFair model can achieve relatively fair recommendations from both the user and item sides while ensuring recommendation performance.}, archive = {J_ASOC}, author = {Chenyu Wang and Guanxi Wang and Guowei Yang and Dun Li}, doi = {10.1016/j.asoc.2025.113862}, journal = {Applied Soft Computing}, month = {12}, pages = {113862}, shortjournal = {Appl. Soft. Comput.}, title = {Causal intervention for feedback information in fairness recommendation}, volume = {184}, year = {2025}, }
• (2025). Fixing deep early exit ensembles for sensor-based human activity recognition through uncertainty quantification. ASOC, 184, 113861. (www)

  @article{ , abstract = {Dynamic deep neural networks (DNNs) have recently achieved outstanding success on a wide range of resource-constrained human activity recognition (HAR) tasks. However, prior most works are deterministic and could not provide any uncertainty estimate. Though an early exit can facilitate adaptive activity inference by producing intermediate predictions at multiple stages during forward pass, these predictions are only meaningful in real-world while complemented with reliable uncertainty estimates. Until now, the quality of uncertainty estimates has always been ignored in the context of early exit HAR. How to quantify predictive uncertainty in dynamic DNNs still remains challenging and yet unsolved. To address this issue, this paper introduces a new framework of early exit ensembles, which provides a probabilistic treatment of such dynamic DNNs to capture uncertainty estimates through an implicit ensemble of sub-networks sharing weights. We evaluate the proposed approach using three strong state-of-the-art DNN backbones on several mainstream HAR benchmarks, i.e., UCI-HAR, UniMib-SHAR and WISDM. Depending on the backbones and datasets, our approach can lower calibration error up to around 11 × , while increasing accuracy by up to 0.588% over its single counterpart. Both theoretical computational efficiency and practical runtime latency are analyzed. We provide an intuitive illustration of accounting for both aleatoric and epistemic uncertainty, which validates that such probabilistic treatment can adequately capture uncertainty estimates to aid decision-making while varying the computational budgets.}, archive = {J_ASOC}, author = {Xin Liu and Lei Zhang and Wenbo Huang and Dongzhou Cheng and Hao Wu and Aiguo Song}, doi = {10.1016/j.asoc.2025.113861}, journal = {Applied Soft Computing}, month = {12}, pages = {113861}, shortjournal = {Appl. Soft. Comput.}, title = {Fixing deep early exit ensembles for sensor-based human activity recognition through uncertainty quantification}, volume = {184}, year = {2025}, }
• (2025). A transformer-based water quality prediction model with mixed fractional brownian features and multi-feature bottleneck transformation. ASOC, 184, 113860. (www)

  @article{ , abstract = {Data-driven water-quality model inference typically depends on large datasets. In emergency or resource-constrained settings, however, small sample sizes hinder a model’s ability to faithfully capture the complex nonlinear and multiscale dynamics of water quality. To address this challenge, we propose an improved Transformer. First, we enrich the inputs with features from mixed fractional Brownian motion with a perturbation factor (rMFBM). The rMFBM module captures heterogeneous temporal dependencies via multiple Hurst exponents and applies a Cholesky-based covariance sampler with a diagonal perturbation to ensure numerical stability. Second, a multi-feature bottleneck layer performs compression and dimensionality reduction to yield compact yet information-dense representations. Finally, the optimized features are modeled by a lightweight Transformer trained with the AMSGrad optimizer and an early-stopping strategy. On the task of predicting the N S F W Q I water-quality index, the proposed model reduces root-mean-square error (RMSE) by approximately 43.1% relative to a standard Transformer and improves the coefficient of determination ( R 2 ) by 1.89% across four small-sample datasets. It also outperforms widely used alternatives, including GNNs and PINNs. These results suggest that certain water-quality parameters exhibit mixed fractional Brownian-motion characteristics, enabling rapid and reliable prediction under data scarcity and in emergency scenarios.}, archive = {J_ASOC}, author = {Genghao Cui and Zhiyao Zhao and Li Wang and Huiyan Zhang and Jiabin Yu}, doi = {10.1016/j.asoc.2025.113860}, journal = {Applied Soft Computing}, month = {12}, pages = {113860}, shortjournal = {Appl. Soft. Comput.}, title = {A transformer-based water quality prediction model with mixed fractional brownian features and multi-feature bottleneck transformation}, volume = {184}, year = {2025}, }
• (2025). MPCMamba: Privacy-preserving inference for mamba models via secure multi-party computation. ASOC, 184, 113859. (www)

  @article{ , abstract = {Deep learning is increasingly critical across high-stakes domains, but privacy risks during model inference remain a major concern. Secure multi-party computation (SMPC) offers a promising solution by enabling privacy-preserving collaborative inference without exposing sensitive data. The recently proposed Mamba model, which outperforms Transformer in certain tasks, presents unique challenges for SMPC due to its state-space architecture and nonlinear operations. This paper introduces a framework for executing Mamba model inference under SMPC while preserving privacy. The framework natively supports linear operations and securely computes nonlinear functions – including square roots, exponentials, logarithms, and SiLU activations – without altering the original model architecture. Experimental results demonstrate that the proposed method achieves accuracy improvements of 2.4%, 2.84%, and 8.23% on CIFAR-10, CIFAR-100, and Tiny-ImageNet datasets, respectively, compared to existing SMPC-based vision Transformer approaches, MPCViT. Additionally, inference latency is reduced by factors of 2.14 × , 2.14 × , and 26.13 × on these benchmarks, significantly advancing efficient and secure deployment of state-space models in privacy-sensitive scenarios.}, archive = {J_ASOC}, author = {Yongqiang Yu and Yuliang Lu and Xuehu Yan and Wei Yan and Shengyang Luo}, doi = {10.1016/j.asoc.2025.113859}, journal = {Applied Soft Computing}, month = {12}, pages = {113859}, shortjournal = {Appl. Soft. Comput.}, title = {MPCMamba: Privacy-preserving inference for mamba models via secure multi-party computation}, volume = {184}, year = {2025}, }
• (2025). A depth-estimation-based method for multi-view synthesis applied to chinese landscape paintings. ASOC, 184, 113858. (www)

  @article{ , abstract = {This paper addresses the challenging task of novel view synthesis for traditional Chinese landscape paintings, which typically offer only a single perspective and lack clear depth information. To overcome the limitations of existing methods that rely on multi-view input and depth estimation, we propose a multi-view synthesis method for Chinese landscape paintings, termed MVSM-CLP. The proposed CLPDepth Module employs a high-low resolution fusion mechanism to enhance detail expression while preserving the original scene structure. We introduce an image restoration technique guided by landscape ink lines, termed LInpainting, to improve edge extraction and the accuracy of painting inpainting Additionally, our method tackles the issue of scarce 3D data in current view synthesis efforts by constructing multi-view data from a single ancient painting. Our approach effectively bridges the gap between 2D art and 3D visualization, creating vivid and realistic virtual environments while preserving the traditional style and essence of Chinese paintings. Experimental results demonstrate the effectiveness of our method in achieving high-quality multi-view synthesis, offering new possibilities for the digital preservation of cultural heritage. A preprint has previously been published (Peng et al., 2024 [1] ). The code and dataset is available at https://github.com/LPDLG/DepCLP_Dataset .}, archive = {J_ASOC}, author = {Xianlin Peng and Wanlin Zhou and Qiyao Hu and Tengfei Li and Dong Zhang and Rui Cao}, doi = {10.1016/j.asoc.2025.113858}, journal = {Applied Soft Computing}, month = {12}, pages = {113858}, shortjournal = {Appl. Soft. Comput.}, title = {A depth-estimation-based method for multi-view synthesis applied to chinese landscape paintings}, volume = {184}, year = {2025}, }
• (2025). SGBA: Subspace guidance backdoor attack with feature alignment in image classification. ASOC, 184, 113857. (www)

  @article{ , abstract = {With the widely use of deep neural networks (DNNs) in a variety of classification and generation tasks with remarkable performance, the security of DNN models has attracted further attention. Recently, backdoor attacks have made a significant threat to DNN models, where attackers intentionally introduce malicious patterns into the model and manipulate the behavior of the model upon encountering the specific trigger. In this paper, we analyze the workflow of backdoor attacks with feature-guide classifier, and propose a novel method to conduct the backdoor attack based on feature alignment. We find that using samples with adversarial perturbation for training can mislead the model during the inference stage. Inspired by this observation, we leverage the intrinsic distribution of target class features in the latent space to generate effective and invisible triggers by adding directional perturbations to target images. We evaluate the attack under three widely used datasets and results show that our method can achieve considerable performance in comparison with other four state-of-the-art attacks. Furthermore, we also make extensive experiment to emphasize the robustness and stealthiness of our attack method.}, archive = {J_ASOC}, author = {Hao Luo and Zhi Qin and Lin Wang and Ziyue Wu and Min Yang}, doi = {10.1016/j.asoc.2025.113857}, journal = {Applied Soft Computing}, month = {12}, pages = {113857}, shortjournal = {Appl. Soft. Comput.}, title = {SGBA: Subspace guidance backdoor attack with feature alignment in image classification}, volume = {184}, year = {2025}, }
• (2025). Deep learning-based kiwifruit flower recognition method to facilitate automated pollination. ASOC, 184, 113855. (www)

  @article{ , abstract = {Accurate flowering-stage recognition is vital for intelligent orchard management, automated pollination, and yield forecasting. However, in complex natural scenes, existing models struggle to balance detection accuracy with inference speed. To bridge this gap, we propose the Kiwifruit Recognition Network (KiwiRecNet), a lightweight yet high-performance framework tailored to kiwifruit blossoms in the wild. KiwiRecNet first employs the Kiwifruit Generative Adversarial Network for Low-light Improvement (KiwiGAN-LI) to enhance under-exposed images. We then design a novel backbone, the Multi-Scale Shuffle Block (MSBlock), which combines structural re-parameterisation with channel–spatial shuffling to shrink the network footprint. Next, we propose the Partial-Mixing Vision Transformer (PMVIT), a convolution-Transformer hybrid that captures fine-grained features and remains robust to occlusion. Finally, we devise a Bidirectional Cross-Scale Fusion module (Bi-CSF) to enrich multiscale perception. Evaluated on the NWAFU Kiwifruit_F dataset, KiwiRecNet achieves 94.07 % mAP at 82.67 FPS with only 0.93 million parameters, outperforming existing lightweight detectors while approaching heavyweight baselines at a fraction of their cost. Consistent gains across multiple flower datasets confirm its generalisation ability. These results demonstrate an effective route to high-accuracy, real-time flowering-phase recognition on resource-constrained devices, paving the way for scalable agricultural automation.}, archive = {J_ASOC}, author = {Xiaopeng Li and Jinzhi Du and Xiaoyu Chen and Fuxi Shi and Shuqin Li}, doi = {10.1016/j.asoc.2025.113855}, journal = {Applied Soft Computing}, month = {12}, pages = {113855}, shortjournal = {Appl. Soft. Comput.}, title = {Deep learning-based kiwifruit flower recognition method to facilitate automated pollination}, volume = {184}, year = {2025}, }
• (2025). Searching for local pareto fronts based on the non-dominance range in the decision space. ASOC, 184, 113853. (www)

  @article{ , abstract = {Multi-modal multi-objective optimisation problems (MMOPs) involve multiple equivalent Pareto sets that share the same Pareto front, and a special subclass is known as MMOPs with local Pareto fronts (MMOPLs). Conventional multi-modal multi-objective optimisation evolutionary algorithms (MMOEAs) struggle to effectively identify local Pareto fronts, and existing methods tailored for MMOPLs exhibit notable limitations. Additionally, commonly used performance metrics lack systematic evaluation and suffer from inherent shortcomings. A simple yet effective MMOEA that leverages the non-dominance range in the decision space is proposed to address these challenges. By prioritising individuals with above-average non-dominance ranges, the algorithm enhances the identification and retention of both global and local optima. Convergence is further improved using the local outlier factor method. The limitations of existing performance metrics are analysed and six new performance metrics tailored for MMOPLs are introduced. To facilitate evaluation, benchmark problems are modified to create scenarios in which global and local optima coexist on the same Pareto set. Extensive experimental results confirm that the proposed algorithm achieves competitive performance, effectively identifying global and local optima while ensuring well-distributed solutions.}, archive = {J_ASOC}, author = {Yimin Shen and Yu Guo and Shaohua Huang and Weiwei Qian and Shengbo Wang and Litong Zhang}, doi = {10.1016/j.asoc.2025.113853}, journal = {Applied Soft Computing}, month = {12}, pages = {113853}, shortjournal = {Appl. Soft. Comput.}, title = {Searching for local pareto fronts based on the non-dominance range in the decision space}, volume = {184}, year = {2025}, }
• (2025). Normalizing flow defect detection model based on similar self-supervision. ASOC, 184, 113847. (www)

  @article{ , abstract = {To overcome the limitations in unsupervised industrial anomaly detection, such as poor modeling of the appearance differences between synthetic and real defects and the failure to capture multi-scale feature relationships, this paper proposes a Normalizing Flow Defect Detection Model Based on Similar Self-Supervision (NF-SS). First, a feature-level defect generation method is introduced. It synthesizes realistic defect samples by fusing defect background features, known defect features from related domains, and other auxiliary features. This enhances the model's understanding of defect characteristics and reduces the gap between synthetic and real defects. Second, a multi-scale joint Normalizing Flow decoder is proposed. It replaces independent NF layers with progressive multi-scale fusion, allowing the model to integrate features hierarchically across scales. This preserves spatial relationships, reduces edge blurring, and improves defect localization accuracy. Extensive experiments on the MVTec AD dataset show that NF-SS outperforms state-of-the-art models, achieving an average image-level AUC of 99.54 % and pixel-level AUC of 98.76 %. It significantly improves the detection of subtle and ambiguous defects. NF-SS combines unsupervised and self-supervised learning, providing a robust solution for industrial quality inspection with limited defect samples.}, archive = {J_ASOC}, author = {Zhenlian Miao and Guangzhu Chen and Herui Cao and Yuan Tang and Xiaojuan Liao}, doi = {10.1016/j.asoc.2025.113847}, journal = {Applied Soft Computing}, month = {12}, pages = {113847}, shortjournal = {Appl. Soft. Comput.}, title = {Normalizing flow defect detection model based on similar self-supervision}, volume = {184}, year = {2025}, }
• (2025). Refined vision to obtain the vibration trajectory of rotating body. ASOC, 184, 113845. (www)

  @article{ , abstract = {Non-contact vision technology has been widely used in many fields, but it still has limited accuracy in dealing with complex motion, small displacement and boundary recognition, especially in the rotating body scene, where it is difficult to balance global robustness and local sensitivity. To achieve high-precision visual measurement of rotating structure vibration, this paper proposes a detection-guided target tracking method to address the shortcomings of existing visual algorithms in target identity preservation and temporal consistency. In the face of the above challenges, this paper constructs lightweight encoder–decoder network to extract multi-scale semantic information to enhance the target edge modeling capability, and designs a multidimensional regression mechanism to predict the target boundary from the center point and extract sub-pixel vibration signals. At the same time, a guide term is introduced to optimize trajectory association and improve the temporal continuity and spatial consistency of tracking. Experimental results show that the average performance of the proposed method is improved by about 51.2% on the basis of multiple mainstream visual algorithms, and the main frequency recognition error is reduced to 0.32 Hz, which effectively suppresses high-frequency vibration errors. Ablation experiments verify the contribution of each module to the measurement accuracy, and low-quality image tests further show that the method has good generalization ability and application prospects.}, archive = {J_ASOC}, author = {Rongliang Yang and Tao Liu and Sen Wang}, doi = {10.1016/j.asoc.2025.113845}, journal = {Applied Soft Computing}, month = {12}, pages = {113845}, shortjournal = {Appl. Soft. Comput.}, title = {Refined vision to obtain the vibration trajectory of rotating body}, volume = {184}, year = {2025}, }
• (2025). Improving lightweight semi-supervised text classification via teacher intervention. ASOC, 184, 113844. (www)

  @article{ , abstract = {The most recent semi-supervised text classification frameworks have focused on pre-trained models like BERT. While effective, their large-scale parameters and slow inference speed hinder deployment in many practical scenarios. In this work, we develop a general Li ghtweight S emi-supervised T ext classification framework (LiST), which significantly enhances the performance of lightweight models in semi-supervised settings, thus improving inference efficiency. LiST employs a teacher intervention strategy, where pseudo-labels initially rely on the teacher model and gradually shift to the lightweight model’s own predictions, progressively correcting the teacher’s incorrect predictions over time. Experimental results on multiple benchmark datasets demonstrate the generality and effectiveness of LiST. It not only approaches or exceeds the performance of the teacher model (UDA) at a 20 × faster inference speed but also outperforms existing lightweight methods in both performance and efficiency. Notably, LiST achieves significant performance improvements even with extremely few labeled samples, such as a 36.4% increase in accuracy on AG News with only 2 labeled samples per class.}, archive = {J_ASOC}, author = {Shaohuan Cheng and Wenyu Chen and Wanlong Liu and Hong Qu}, doi = {10.1016/j.asoc.2025.113844}, journal = {Applied Soft Computing}, month = {12}, pages = {113844}, shortjournal = {Appl. Soft. Comput.}, title = {Improving lightweight semi-supervised text classification via teacher intervention}, volume = {184}, year = {2025}, }
• (2025). Chaotic property based multi-interval informer modeling method for long-term photovoltaic power generation prediction. ASOC, 184, 113843. (www)

  @article{ , abstract = {The long-term prediction of photovoltaic (PV) power generation capacity can significantly enhance the maintenance planning of photovoltaic power stations and support the long-term development strategies of power supply and distribution networks. Currently, two major challenges hinder the realization of effective long-term prediction for PV power generation: First, PV power generation data is heavily influenced by environmental factors, leading to high randomness and significant volatility in long-term data sequences; Second, the temporal continuity of long-term data must be considered, which complicates the construction of an accurate predictive model. To address these issues, this paper proposes a long-term PV power generation prediction method based on interval division, modeled using an improved Informer architecture with a modified activation function. Initially, by analyzing the fluctuation characteristics of PV power generation data, the K-nearest neighbors algorithm is applied for data interpolation, while the moving average (MA) method is employed for data smoothing, effectively reducing data randomness while preserving the overall trend. Subsequently, the Kmeans＋＋ clustering algorithm is utilized to group the generation data into multiple intervals, thereby enhancing feature similarity within each cluster. Finally, the activation function of the Informer model is replaced with ReLU to improve its adaptability to photovoltaic data characteristics. Additionally, this study introduces evaluation metrics such as the Hurst exponent and the Maximum Lyapunov Exponent (MLE) to assess the predictability of partitioned data and identify the interval to which predicted values belong. Simulation experiments demonstrate that, compared to several commonly used prediction models, the proposed interval-based approach achieves superior performance in long-term PV power generation prediction.}, archive = {J_ASOC}, author = {Ying Han and Xinggang Hu and Kun Li}, doi = {10.1016/j.asoc.2025.113843}, journal = {Applied Soft Computing}, month = {12}, pages = {113843}, shortjournal = {Appl. Soft. Comput.}, title = {Chaotic property based multi-interval informer modeling method for long-term photovoltaic power generation prediction}, volume = {184}, year = {2025}, }
• (2025). A noisy multi-objective evolutionary optimization algorithm based on elman neural network. ASOC, 184, 113842. (www)

  @article{ , abstract = {Neural networks have emerged as potent tools for tackling multi-objective optimization problems with remarkable success. This research delves into the performance of dynamic neural networks in resolving noisy multi-objective optimization problems (NMOPs). The local regression structure of the Elman neural network enables it to deal with the dynamic noise problem well. This paper integrates the Elman neural network into the framework of the non-dominated sorting genetic algorithm II (NSGA-II) to solve NMOP, called E-NSGA-II. In this method, the Elman neural network is employed for modeling, estimating the fitness value of each individual. Simultaneously, a hybrid selection solution strategy is implemented to select individuals for modeling, providing more diverse solutions in the modeling process to improve the convergence. Furthermore, E-NSGA-II incorporates a noise-driven sampling mechanism that intelligently adapts the sampling frequency based on the noise intensity. This feature could not only enhance the accuracy of neural network modeling, but also minimize the amount of calculation. Notably, the embedding of neural network does not impose much additional evaluation overheads, thereby bolstering the overall efficiency of E-NSGA-II. Experimental results prove the superiority of E-NSGA-II over four state-of-the-art noisy multi-objective optimization algorithms on dealing with NMOPs.}, archive = {J_ASOC}, author = {Jianxia Li and Ruochen Liu and Wanfeng Chen and Weibin Li}, doi = {10.1016/j.asoc.2025.113842}, journal = {Applied Soft Computing}, month = {12}, pages = {113842}, shortjournal = {Appl. Soft. Comput.}, title = {A noisy multi-objective evolutionary optimization algorithm based on elman neural network}, volume = {184}, year = {2025}, }
• (2025). An effective detection algorithm for small UAV based on lightweight you-only-look-once (YOLOv4-l) approach. ASOC, 184, 113841. (www)

  @article{ , abstract = {The control and monitoring of small Unmanned Aerial Vehicles (UAV) plays a crucial role in national defense and security. However, due to their compact size and high mobility, the detection of small UAV across diverse scenarios remains a significant challenge. To address this issue, this study proposes an improved detection algorithm tailored for small UAV. The model is initially trained on a virtual dataset, and the learned parameters are transferred to real-world data through a transfer learning framework. To optimize anchor box generation, clustering analysis is performed on bounding box dimensions, resulting in anchor boxes with appropriate scales and aspect ratios. Furthermore, the Ghost module is introduced to replace conventional convolutions in CSPDarknet53, enhancing feature extraction efficiency. An Efficient Channel Attention (ECA) mechanism is also incorporated to strengthen output feature representations and improve the capture of texture details critical for small target detection. Through experiments, the proposed algorithm can achieve the mAP0.5 of 82.2 %. Experimental results demonstrate the effectiveness of the proposed small UAV detection method.}, archive = {J_ASOC}, author = {Guoning Li and Jianghao Cheng and Yanyan Liu and Jin Li and Zengming Lv and Qiang Li}, doi = {10.1016/j.asoc.2025.113841}, journal = {Applied Soft Computing}, month = {12}, pages = {113841}, shortjournal = {Appl. Soft. Comput.}, title = {An effective detection algorithm for small UAV based on lightweight you-only-look-once (YOLOv4-l) approach}, volume = {184}, year = {2025}, }
• (2025). TPLLM: A traffic prediction framework based on pretrained large language models. ASOC, 184, 113840. (www)

  @article{ , abstract = {Traffic prediction constitutes a critical component in sustainable urban data analysis, playing a pivotal role in optimizing transportation systems for reduced carbon emissions and improved energy efficiency. The precision of prevailing deep learning-driven traffic prediction models typically improves as the volume of training data increases. However, the procurement of comprehensive spatiotemporal datasets for traffic is often fraught with challenges, primarily stemming from the substantial costs associated with data collection and retention. This limitation severely hinders the deployment of models in regions with insufficient historical data. Consequently, developing a model that can achieve accurate predictions and good generalization ability in areas with limited historical traffic data is a challenging problem. It is noteworthy that the rapidly advancing pretrained Large Language Models (LLMs) of recent years demonstrate exceptional proficiency in cross-modality knowledge transfer and few-shot learning. Recognizing the sequential nature of traffic data, similar to language, we introduce TPLLM, a novel traffic prediction framework leveraging LLMs. In this framework, we construct a sequence embedding layer based on Convolutional Neural Networks (CNNs) and a graph embedding layer based on Graph Convolutional Networks (GCNs) to extract sequence features and spatial features, respectively. These are subsequently integrated to form inputs that are suitable for LLMs. A Low-Rank Adaptation (LoRA) fine-tuning approach is applied to TPLLM, thereby facilitating efficient learning and minimizing computational demands. Experiments on two real-world datasets demonstrate that TPLLM exhibits commendable performance in both full-sample and few-shot prediction scenarios.}, archive = {J_ASOC}, author = {Tian Ma and Yixuan Zhao and Minda Li and Yue Chen and Fangshu Lei and Yanan Zhao and Maazen Alsabaan}, doi = {10.1016/j.asoc.2025.113840}, journal = {Applied Soft Computing}, month = {12}, pages = {113840}, shortjournal = {Appl. Soft. Comput.}, title = {TPLLM: A traffic prediction framework based on pretrained large language models}, volume = {184}, year = {2025}, }
• (2025). Dynamic-linguistic fusion for brain tumor classification: A cross-modal attention framework with clinical interpretability. ASOC, 184, 113839. (www)

  @article{ , abstract = {Accurate classification of brain tumors is a key challenge in medical image analysis, and existing methods mainly rely on static MRI images, which are difficult to capture the dynamic evolutionary features of tumors. In addition, the lack of descriptive clinical text and efficient multimodal feature fusion limits the classification accuracy and generalization. To overcome these limitations, this paper proposes a new dynamic language fusion (DLF) framework. The framework (1) utilizes ResNet18 in conjunction with LSTM for time series modeling to capture the temporal evolution of tumor morphology, (2) uses BioGPT and BERT for clinical text processing for semantic understanding, and (3) applies an interpretable cross-modal attentional mechanism for feature fusion to optimize dynamic perception and semantic alignment. Experiments on 10,287 images (from four publicly available datasets) show that the proposed framework achieves an overall accuracy of 98.96 %, a precision of 99.58 %, and an AUC higher than 0.998 for all categories on the test set, which is significantly better than the existing SOTA models, and especially exhibits stronger robustness and discriminative ability in boundary ambiguity and feature overlap samples. This study validates the synergistic effect of temporal modeling and semantic understanding in brain tumor diagnosis, providing clinicians with interpretable classification outputs to assist in decision-making for complex cases, while establishing a scalable framework for medical AI systems based on large language models.}, archive = {J_ASOC}, author = {Jiancong Fan and Fangyuan Chen and Yang Li and Jiehan Zhou}, doi = {10.1016/j.asoc.2025.113839}, journal = {Applied Soft Computing}, month = {12}, pages = {113839}, shortjournal = {Appl. Soft. Comput.}, title = {Dynamic-linguistic fusion for brain tumor classification: A cross-modal attention framework with clinical interpretability}, volume = {184}, year = {2025}, }
• (2025). Path planning for flexible needle puncture based on multi-objective particle swarm optimization. ASOC, 184, 113838. (www)

  @article{ , abstract = {This paper presents a mixed-parameter MOPSO algorithm designed to address the puncture problem of flexible needles in obstacle environments. The algorithm incorporates the damage caused by the pivotal angle to soft tissues as an objective function, marking the first time this has been applied to the MOPSO algorithm. In comparison with four classical algorithms through simulation experiments, the path deviation is reduced to just 0.1 mm, significantly lower than the CPSO algorithm. The final path score achieves 35 points, surpassing the performance of other algorithms. A self-built FPAA hybrid control platform was employed for puncture experiments using a gelatin prosthesis. The experimental results confirm that the flexible needle successfully avoids obstacles and reaches the target, demonstrating the feasibility of the proposed puncture algorithm.}, archive = {J_ASOC}, author = {Ting Yang and Beibei Liu and Ru Sun and Bi Chen and Guijuan Ji}, doi = {10.1016/j.asoc.2025.113838}, journal = {Applied Soft Computing}, month = {12}, pages = {113838}, shortjournal = {Appl. Soft. Comput.}, title = {Path planning for flexible needle puncture based on multi-objective particle swarm optimization}, volume = {184}, year = {2025}, }
• (2025). A multi-goal reinforcement learning framework for motion planning of a quadrotor UAV in 3D cluttered environment with unseen random goals. ASOC, 184, 113836. (www)

  @article{ , abstract = {The standard reinforcement learning (RL) approaches render an optimal solution for a specific task by learning an optimal policy. Therefore, these RL approaches are suitable for single goal problems only. However, there exist numerous practical challenges that involve multiple goals to be achieved in a specific environment. For example, navigation of an unmanned aerial vehicle (UAV) to achieve multiple random targets in a specific three-dimensional (3D) cluttered space is a multi-goal problem. Using the standard RL approaches, the UAV agent needs to learn a separate optimal policy for each target. Thus, a multi-goal problem with random unseen goal allocations, especially in the 3D space of UAVs, increases computational effort substantially. To solve this issue, this paper presents a relatively generalized approach that learns the optimal state values and related optimal policies considering various regions of the robot environment. The proposed approach transforms the 3D robot environment into a Markov decision process (MDP) and further divides it into various virtual sub-spaces. For every initial goal position in a sub-space, value iteration algorithm enables the aerial agent to learn the optimal state values and the relevant policy. These optimal state values and policies are then stored in a replay buffer for later use. Once the initial state values for every sub-space are learned, the proposed approach allows the agent to use them from replay buffer and run a local search algorithm to connect every new goal position to any feasible state in the existing state values. Our proposed framework significantly reduces the computational effort for multiple unseen goal targets by restricting the re-computation of state values for each new goal. To validate the proposed method, a simulator is designed and an RL-based motion planning approach for a quadrotor UAV is presented. The results under various scenarios confer the superior performance of our anticipated multi-goal RL framework.}, archive = {J_ASOC}, author = {Ghulam Farid and Lanyong Zhang and Talha Younas and Muhammad Ilyas and Asma Iqbal}, doi = {10.1016/j.asoc.2025.113836}, journal = {Applied Soft Computing}, month = {12}, pages = {113836}, shortjournal = {Appl. Soft. Comput.}, title = {A multi-goal reinforcement learning framework for motion planning of a quadrotor UAV in 3D cluttered environment with unseen random goals}, volume = {184}, year = {2025}, }
• (2025). Wind speed prediction based on U-shaped 2D multi-scale model. ASOC, 184, 113835. (www)

  @article{ , abstract = {Accurate wind speed forecasting is essential for enhancing energy efficiency and reducing maintenance costs in wind power systems. However, existing CNN-LSTM and Transformer-based models face limitations in capturing periodic features, modeling long-term dependencies, and preserving information during multi-layer encoding. To overcome these challenges, this paper proposes a novel U-shaped 2D multi-scale model that integrates Bi-RLSTM encoding, wavelet transform, and multi-scale 2D convolution. The proposed model first encodes input sequences into a high-dimensional space via feature embedding and Bi-RLSTM, effectively capturing long-term dependencies. Subsequently, a wavelet transform extracts primary fluctuation patterns and their periods, converting 1D sequences into 2D feature maps to enhance periodic feature representation while preserving temporal information. Multi-scale convolutional layers are then employed to extract fine-grained spatial–temporal features from these maps. Finally, a Bi-RLSTM decoding layer, augmented with skip connections, mitigates information loss during deep encoding and reinforces long-range correlation modeling. Extensive experiments on two real-world wind speed datasets demonstrate that the proposed model significantly outperforms CNN-LSTM and attention-based approaches achieving superior performance in terms of MAE, RMSE, and R 2 , confirming its effectiveness and application potential.}, archive = {J_ASOC}, author = {Yue Gao and Zhongda Tian}, doi = {10.1016/j.asoc.2025.113835}, journal = {Applied Soft Computing}, month = {12}, pages = {113835}, shortjournal = {Appl. Soft. Comput.}, title = {Wind speed prediction based on U-shaped 2D multi-scale model}, volume = {184}, year = {2025}, }
• (2025). Enhancing multi-task learning-based tornado identification using spatial and temporal information from weather radar images. ASOC, 184, 113834. (www)

  @article{ , abstract = {Tornadoes, as dynamic weather phenomena, exhibit unique spatial and temporal evolution characteristics that reflect their formation and development. Existing tornado detection algorithms often struggle with high false alarm rates, primarily due to insufficient capture of temporal correlations in tornado development. As an improvement, we propose a multi-task tornado identification network with three-dimensional temporal and spatial information (TS-MTINet). Taking continuous three-frame radar data as input, the Multi-frame Temporal Interaction Block (MTIB) utilizes multi-head attention to model the dynamic interaction information between the radar data, thus exploring in-depth the temporal features during tornado development. Further, we design a Spatial-Temporal Enhancement Module (STEM), which analyzes the difference information between continuous data to extract local and global spatial and temporal feature variations about tornadoes. Based on this architecture, TS-MTINet incorporates a multi-task learning framework to perform tornado detection and number estimation tasks simultaneously, thus extracting comprehensive information related to tornadoes. To validate the performance of the proposed model, we construct the first Chinese tornado identification dataset with fine radar features. The experimental results show that the proposed method shows significant advantages in several evaluation metrics, especially in reducing false alarms. In practical case studies, compared to the traditional TVS method, TS-MTINet achieves an increase in POD of approximately 30% and a decrease in FAR of about 20% in several typical tornado events. Particularly in environments with strong interference, TS-MTINet demonstrates higher detection accuracy, reflecting greater robustness and practical value.}, archive = {J_ASOC}, author = {Jinyang Xie and Kanghui Zhou and Lei Han and Liang Guan and Maoyu Wang and Yongguang Zheng and Hongjin Chen and Jiaqi Mao}, doi = {10.1016/j.asoc.2025.113834}, journal = {Applied Soft Computing}, month = {12}, pages = {113834}, shortjournal = {Appl. Soft. Comput.}, title = {Enhancing multi-task learning-based tornado identification using spatial and temporal information from weather radar images}, volume = {184}, year = {2025}, }
• (2025). Evaluating regional green industry competitiveness in china: A CoSOGR-MABAC-sort framework. ASOC, 184, 113831. (www)

  @article{ , abstract = {With the rise of concepts such as the green economy and sustainable development, the development of green industry has become increasingly urgent. Regional green industry competitiveness, as a key indicator reflecting the development status and competitive advantages of regional green industry, has not yet received sufficient attention and research. To fill this research gap, this paper proposes an evaluation framework, CoSOGR-MABAC-Sort (Combined Subjective-Objective, Grey Relational analysis-Multi-Attributive Border Approximation area Comparison-Sort), to assess the green industry competitiveness in 31 regions of China. In the proposed framework, the MABAC-Sort method is a novel multi-criteria decision sorting method with six classification rules. Compared to other MCDS methods, this method does not require predefined profile boundaries and can provide more detailed classification results. Furthermore, we propose an optimal model , called CoSOGR, to determine the weights of each indicator. Finally, by collecting subjective and objective data, we use the proposed framework to assess the green industry competitiveness in the 31 provinces of China. The main findings are as follows: 1) The CoSOGR demonstrates the highest consistency (99.33 %) in regions ranking compared to existing weight methods (CRITIC, EWM, ROCOSD, and MEREC). 2) The CoSOGR addresses the weight determination issues that ROCOSD (a mixed-integer linear programming model) cannot resolve. 3) The CoSOGR-MABAC-Sort method exhibit strong robustness and stability. 4) The green industry competitiveness in each region shows a positive correlation with the sustainable development of that region.}, archive = {J_ASOC}, author = {Jiafu Su and Baojian Xu and Lvcheng Li and Yijun Chen and Hongyu Liu and Na Zhang}, doi = {10.1016/j.asoc.2025.113831}, journal = {Applied Soft Computing}, month = {12}, pages = {113831}, shortjournal = {Appl. Soft. Comput.}, title = {Evaluating regional green industry competitiveness in china: A CoSOGR-MABAC-sort framework}, volume = {184}, year = {2025}, }
• (2025). AutoTandemML: Active learning enhanced tandem neural networks for inverse design problems. ASOC, 184, 113828. (www)

  @article{ , abstract = {Inverse design in science and engineering involves determining optimal design parameters that achieve desired performance outcomes, a process often hindered by the complexity and high dimensionality of design spaces, leading to significant computational costs. To tackle this challenge, we propose a novel hybrid approach that combines active learning with Tandem Neural Networks to enhance the efficiency and effectiveness of solving inverse design problems. Active learning allows to selectively sample the most informative data points, reducing the required dataset size without compromising accuracy. We investigate this approach using three benchmark problems: airfoil inverse design, photonic surface inverse design, and scalar boundary condition reconstruction in diffusion partial differential equations. We demonstrate that integrating active learning with Tandem Neural Networks outperforms standard approaches across the benchmark suite, achieving better accuracy with fewer training samples.}, archive = {J_ASOC}, author = {Luka Grbcic and Juliane Müller and Wibe Albert de Jong}, doi = {10.1016/j.asoc.2025.113828}, journal = {Applied Soft Computing}, month = {12}, pages = {113828}, shortjournal = {Appl. Soft. Comput.}, title = {AutoTandemML: Active learning enhanced tandem neural networks for inverse design problems}, volume = {184}, year = {2025}, }
• (2025). Personalized cross-domain recommendation with meta networks and contrastive learning. ASOC, 184, 113827. (www)

  @article{ , abstract = {In this paper, we propose a novel cross-domain recommendation (CDR) model that integrates matrix factorization (MF), attention networks, meta-networks, and contrastive learning (CL) to address the challenges of data sparsity and cold-start problems in recommender systems. Our model consists of five modules: the MF module extracts user and item embeddings from rating matrices in both information-rich and information-scarce domains; the transferable feature extraction module uses an attention network to identify and extract transferable features from users’ interactions in the information-rich domain; the cross-domain meta-knowledge transfer module employs a meta-network to transfer these features across domains while preserving users’ personalized preferences; the intra-domain CL module ensures temporal consistency of users’ preferences by learning from their interaction sequences in the information-rich domain; and the inter-domain CL module leverages feedback from users’ interactions in the information-scarce domain to refine the transferable features. We conduct extensive experiments on the Amazon and Douban datasets, evaluating our model across three different CDR tasks. The results demonstrate that our proposed model outperforms other prevalent CDR models in terms of MAE and RMSE. Additionally, our model shows the robust performance in cold-start scenarios, effectively utilizing both intradomain and interdomain knowledge to enhance the recommendation accuracy.}, archive = {J_ASOC}, author = {Shudong Liu and Xiping Hao and Xu Chen and Wenming Ma and Feng Gu}, doi = {10.1016/j.asoc.2025.113827}, journal = {Applied Soft Computing}, month = {12}, pages = {113827}, shortjournal = {Appl. Soft. Comput.}, title = {Personalized cross-domain recommendation with meta networks and contrastive learning}, volume = {184}, year = {2025}, }
• (2025). Efficient real-time visual anomaly detection via frequency-aware diffusion model and information fusion. ASOC, 184, 113826. (www)

  @article{ , abstract = {Real-time visual anomaly detection is always a big challenge since advanced techniques usually suffer from intensive computation, while simple methods cannot obtain desirable performance. To address this issue, an Efficient Real-time Anomaly Detection method via frequency-aware diffusion model and information fusion (ERAD) is proposed, which uses the diffusion model to achieve a nice reconstruction for the frequency information obtained by Discrete Wavelet Transform (DWT) and utilizes the information fusion technique to take full consideration of local and global various features. First, the input image is fed into the DWT module to produce one low-frequency and three high-frequency coefficient images, in this way, data dimensions can be reduced by 75% but the important information can be greatly retained. Then, a diffusion model with one-step denoising is developed rather than the traditional iterative denoising to accelerate the reconstruction speed. As well, information fusion in the framework of selective fusion is embedded into the reconstruction process to improve the model performance. Furthermore, during the segmentation, a wavelet-based upsampling module is put forward to seamlessly combine global semantic context with detailed edge information, achieving both consistent semantics and high-resolution feature reconstruction. Finally, extensive experiments conducted on a variety of benchmark datasets demonstrate the remarkable superiority of our method in both model performance and time efficiency. Specifically, the proposed method achieves an Image AUROC of 99.7 and a Pixel AUROC of 99.5, with an inference time of only 0.04 s.}, archive = {J_ASOC}, author = {Xianzhe Yao and Ping Kong and Quanquan Li and Yan Song}, doi = {10.1016/j.asoc.2025.113826}, journal = {Applied Soft Computing}, month = {12}, pages = {113826}, shortjournal = {Appl. Soft. Comput.}, title = {Efficient real-time visual anomaly detection via frequency-aware diffusion model and information fusion}, volume = {184}, year = {2025}, }
• (2025). Dual-splitting conformal prediction for multi-step time series forecasting. ASOC, 184, 113825. (www)

  @article{ , abstract = {Time series forecasting is crucial for applications like resource scheduling and risk management, where multi-step predictions provide a comprehensive view of future trends. Uncertainty Quantification (UQ) is a mainstream approach for addressing forecasting uncertainties, with Conformal Prediction (CP) gaining attention due to its model-agnostic nature and statistical guarantees. However, most variants of CP are designed for single-step predictions and face challenges in multi-step scenarios, such as reliance on real-time data and limited scalability. This highlights the need for CP methods specifically tailored to multi-step forecasting. We propose the Dual-Splitting Conformal Prediction (DSCP) method, a novel CP approach designed to capture inherent dependencies within time-series data for multi-step forecasting. Experimental results on real-world datasets from four different domains demonstrate that DSCP significantly outperforms existing CP variants in terms of the Winkler Score, improving performance by up to 23.59% compared to state-of-the-art methods. Furthermore, the deployment of DSCP for renewable energy generation and IT load forecasting in the power management of a real-world trajectory-based application achieves an 11.25% reduction in carbon emissions through predictive optimization of data center operations and control strategies.}, archive = {J_ASOC}, author = {Qingdi Yu and Zhiwei Cao and Ruihang Wang and Zhen Yang and Lijun Deng and Min Hu and Yong Luo and Xin Zhou}, doi = {10.1016/j.asoc.2025.113825}, journal = {Applied Soft Computing}, month = {12}, pages = {113825}, shortjournal = {Appl. Soft. Comput.}, title = {Dual-splitting conformal prediction for multi-step time series forecasting}, volume = {184}, year = {2025}, }
• (2025). FDP-net: Fourier transform guided lightweight depthwise and pointwise dynamic pooling based neural network for medical image classification. ASOC, 184, 113824. (www)

  @article{ , abstract = {Nowadays, deep learning-based medical image classification has become essential, especially in developing countries because of the high volume of patients with less medical professionals as well as required infrastructures. Deep learning models often help in the early detection of diseases; however, it require a high amount of processing power, and sometimes it becomes less scalable for various computer-aided diagnosis. To this end, in this paper, a lightweight Fourier Transform guided Depth and Pointwise Dynamic Pooling based Neural Network (FDP-Net), has been proposed for medical image classification. This paper introduces a Depth and Pointwise Feature Fusion (DPFF) block for learning the important features with less computation and without increasing the model parameters. It also proposes a dynamic pooling technique, an alternative to traditional max-pooling, which dynamically selects the important features. The proposed FDP-Net model is trained to classify medical images with the guidance of Fourier Transformation and multitask loss function, which makes the model converge faster and reduces overfitting. The proposed model has been tested on Acute Lymphoblastic Leukemia (ALL) dataset, Peripheral Blood Cell (PBC) dataset, and Raabin White blood Cell (Raabin-WBC) dataset, and it outperforms the state-of-the-art models with 100%, 98.13% and 96.79% classification accuracies, respectively. Additionally, the proposed model is made with only 0.349 million parameters, thereby enabling faster processing. Code will be avilabe at https://github.com/asfakali/FDP-Net .}, archive = {J_ASOC}, author = {Asfak Ali and Rajdeep Pal and Aishik Paul and Ram Sarkar}, doi = {10.1016/j.asoc.2025.113824}, journal = {Applied Soft Computing}, month = {12}, pages = {113824}, shortjournal = {Appl. Soft. Comput.}, title = {FDP-net: Fourier transform guided lightweight depthwise and pointwise dynamic pooling based neural network for medical image classification}, volume = {184}, year = {2025}, }
• (2025). Explicit cooperation mechanism and multimodal fusion prediction model empower DouDizhu agents. ASOC, 184, 113823. (www)

  @article{ , abstract = {Imperfect information games require agents to make decisions under uncertainty, presenting significant challenges yet offering broad applicability in real-world scenarios. DouDizhu, a representative example, features complex cooperative-competitive dynamics, imperfect information, and a large strategy space. To address these challenges, a strategic decision-making algorithm is developed by integrating multi-agent credit allocation with information prediction. Specifically, an explicit credit allocation mechanism based on team reward decomposition improves cooperative behavior among peasant agents by enabling more stable and targeted policy updates. Meanwhile, combining long short-term memory networks and multi-head attention enhances the prediction of hidden opponent information through multimodal feature fusion. Moreover, convolutional neural networks are incorporated into the DouZero framework to extract high-level features and reduce the policy solution space. The resulting CAPRE_DMC significantly improves agent performance in adversarial settings. Extensive evaluations demonstrate that CAPRE_DMC outperforms baseline DouDizhu agents, achieving substantial gains in point margin and win percentage. Additionally, evaluations on benchmark multi-agent cooperative tasks demonstrate the framework’s scalability and generality in large-scale imperfect-information environments.}, archive = {J_ASOC}, author = {Jiao Wang and Longyue Fu and Xiang Li and Hongchen Luo and Zhifen Guo}, doi = {10.1016/j.asoc.2025.113823}, journal = {Applied Soft Computing}, month = {12}, pages = {113823}, shortjournal = {Appl. Soft. Comput.}, title = {Explicit cooperation mechanism and multimodal fusion prediction model empower DouDizhu agents}, volume = {184}, year = {2025}, }
• (2025). A lightweight detection method for X-ray security inspection based on YOLOV8. ASOC, 184, 113822. (www)

  @article{ , abstract = {In the field of public safety, the prohibited items detection during X-ray security inspection is crucial for preventing potential threats. The recent advances in artificial intelligence, particularly deep learning, have achieved success in this area. However, the high number of parameters and computational load of these deep learning-based object detection methods result in significant hardware requirements limiting their practical application. To address these issues, a lightweight method for detecting prohibited items in X-ray security inspections, utilizing the YOLOV8 framework, is proposed in this paper. The key innovations of this method are threefold. First, a lightweight convolution module is designed to optimize the YOLOV8 model. This optimization significantly reduces both the number of parameters and the computational load, making the model more efficient. Second, an adaptive spatial-and-channel attention module is designed to enhance feature extraction capabilities. This module enhances feature extraction capabilities without compromising detection accuracy. Third, the Wise-IOU loss function is incorporated to enhance the overall performance of the detection method during training. Finally, the method is evaluated on our real X-ray pseudo-color image dataset, PIDRAY and CLCXRAY with the existing methods. Our proposed method attains a detection accuracy of 98.83 % on the self-constructed dataset, demonstrating a 0.6 % improvement, while concomitantly reducing the parameters by 64.2 % and computational load by 66.3 %. Furthermore, it achieves accuracy enhancements of 0.7 % and 0.82 % on the two public datasets, respectively. The experimental results indicate that this method not only reduces hardware requirements due to a lower number of parameters and computational load but also broadens the model's applicability. This work underscores the importance of balancing model complexity with performance and sets the stage for future research in AI-driven security inspection technologies.}, archive = {J_ASOC}, author = {Xizhuo Yu and Chunyang Chen and Shu Cheng and Jingming Li}, doi = {10.1016/j.asoc.2025.113822}, journal = {Applied Soft Computing}, month = {12}, pages = {113822}, shortjournal = {Appl. Soft. Comput.}, title = {A lightweight detection method for X-ray security inspection based on YOLOV8}, volume = {184}, year = {2025}, }
• (2025). Probabilistic transformation of basic probability assignment based on weighted visibility graph networks. ASOC, 184, 113821. (www)

  @article{ , abstract = {Dempster-Shafer evidence theory (DSET) provides a powerful framework for uncertain reasoning, offering a theoretical basis for decision-making under ambiguity. However, its core representation, the basic probability assignment (BPA), cannot be directly applied to probabilistic decision-making, prompting the need for effective probability transformation methods. A key challenge lies in quantifying the uncertainty inherent in BPAs to guide this transformation process. To address this, we propose an improved probabilistic transformation method that integrates belief entropy and weighted visibility graph networks, which yields more accurate and interpretable probability distributions than existing approaches. Specifically, given a frame of discernment and mass function, we first apply two refined belief entropy measures to evaluate the informational content of each focal element. Based on these entropy-derived orderings, we construct a weighted visibility graph that captures the structural relationships among focal elements. The weights from this graph are then used to compute a proportional belief transformation. Experimental validation across benchmark cases on classical BPA scenarios demonstrates that our method outperforms traditional approaches in terms of entropy consistency and decision quality, as evidenced by lower Kullback–Leibler (KL) divergence, higher probability information capacity (PIC), and reduced Shannon entropy. These results highlight the method’s dual advantage in balancing decisiveness (via PIC maximization) and fidelity (via entropy minimization), making it a robust tool for uncertainty-aware decision support systems.}, archive = {J_ASOC}, author = {Yongchuan Tang and Kangkang Wu and Rongfei Li and He Guan and Deyun Zhou and Yubo Huang}, doi = {10.1016/j.asoc.2025.113821}, journal = {Applied Soft Computing}, month = {12}, pages = {113821}, shortjournal = {Appl. Soft. Comput.}, title = {Probabilistic transformation of basic probability assignment based on weighted visibility graph networks}, volume = {184}, year = {2025}, }
• (2025). Electric network frequency-based digital audio tampering event identification using multimodal feature interaction network. ASOC, 184, 113820. (www)

  @article{ , abstract = {In the field of digital Audio Tampering Event (ATE) detection based on Electric Network Frequency (ENF), accurately extracting ENF signals is crucial for tampering event identification. However, ENF signals are susceptible to noise interference, and it is difficult to establish an effective matching relationship with the reference frequency database, which poses a challenge to the effectiveness of existing ATE detection methods. To solve these problems, a Multimodal Feature Interactive Network (MFIN) is proposed for ATE identification under low-SNR conditions. First, a Modified Kaiser-window-based S-Transform (MKST) method is proposed to extract the estimated frequency, phase, rate of change of frequency, and rate of change of phase of the ENF component in digital audio. The frequency and time resolution are increased through the improved control function, thereby enhancing the accuracy of ENF estimation. Subsequently, a Multi-Branch Multi-Scale Attention Convolutional (MBSAC) neural network is further proposed to extract and classify ENF features. In MBSAC, the Multi-Branch Multi-Scale Temporal (MBST) attention fusion block, serving as the network’s primary structure, increases feature diversity and enhances classification performance. Finally, the ATE data acquisition hardware platform is built. Experimental results on a dataset comprising six types of ATEs demonstrate that the proposed MFIN outperforms several state-of-the-art ATE detection methods in both ENF extraction and tampering type classification.}, archive = {J_ASOC}, author = {Bing Li and Junfeng Duan and Yao Zheng and Xinxin Cai and Wei Qiu and He Yin and Wenxuan Yao}, doi = {10.1016/j.asoc.2025.113820}, journal = {Applied Soft Computing}, month = {12}, pages = {113820}, shortjournal = {Appl. Soft. Comput.}, title = {Electric network frequency-based digital audio tampering event identification using multimodal feature interaction network}, volume = {184}, year = {2025}, }
• (2025). Enhancing financial resilience in manufacturing SMEs: A q-rung picture fuzzy set-based decision framework for digital transformation adoption. ASOC, 184, 113819. (www)

  @article{ , abstract = {Digital transformation (DT) has emerged as a crucial strategy for enhancing the financial resilience of small and medium-sized enterprises (SMEs) in the manufacturing sector. Despite its importance, there is a significant gap in the quantitative evaluation of the factors influencing SMEs’ DT adoption. This study addresses this gap by developing a decision framework using q-rung picture fuzzy sets (q-RPF) to identify and analyze the drivers of DT implementation aimed at boosting financial resilience in manufacturing SMEs. The proposed framework incorporates the q-RPF-weighted Heronian mean aggregation operator to aggregate expert evaluations and capture the interrelationships among input decision data. A novel weighting approach, based on the q-RPF deviation measure, is introduced to assess the significance of various enablers. Furthermore, the q-RPF-MARCOS’H model is employed to evaluate the effectiveness of DT in enhancing financial resilience across different SMEs by integrating the methodologies mentioned above. A numerical example involving manufacturing SMEs from a specific city demonstrates the practical application of the q-RPF-MARCOS’H model-based framework. The results reveal that “concurrent operations” (0.102) is a significant enabler of DT adoption in promoting financial resilience. The framework’s validity is confirmed through both sensitivity and comparative analyses. These outcomes offer recommendations for policymakers and industry leaders to design effective incentives for DT adoption and provide practical guidance for SMEs seeking to leverage DT for improved financial resilience.}, archive = {J_ASOC}, author = {Zhengyan Yang and Wei Zhong Wang and Zelin Wang and Muhammet Deveci and Dursun Delen}, doi = {10.1016/j.asoc.2025.113819}, journal = {Applied Soft Computing}, month = {12}, pages = {113819}, shortjournal = {Appl. Soft. Comput.}, title = {Enhancing financial resilience in manufacturing SMEs: A q-rung picture fuzzy set-based decision framework for digital transformation adoption}, volume = {184}, year = {2025}, }
• (2025). A regret theory-based three-way decision model under comparative linguistic expressions. ASOC, 184, 113816. (www)

  @article{ , abstract = {In real world, experts often prefer to utilize linguistic expressions over numerical data when evaluating alternatives. However, due to the complexity of actual decision-making, single linguistic terms are insufficient for experts to express their judgments accurately. Thus, it is necessary to employ a richer form of linguistic expression, known as comparative linguistic expressions (CLEs). Furthermore, existing multi-attribute decision-making (MADM) models under CLE suffer from the following limitations: (1) they fail to provide decision-making references for alternatives; (2) they do not incorporate the psychological factors of experts. In order to address the aforementioned challenges, this paper proposes a novel regret theory (RT)-based three-way decision (TWD) model under CLE. First, the attribute weights are calculated by an improved optimization model. This model combines index variability with comprehensive entropy, enabling a more objective conclusion to be drawn regarding the relative importance of attributes. Second, an enhanced neighborhood relationship is introduced, which is shown to fulfill the properties of Symmetry, Reflexivity, and Non-transitivity. Building on this foundation, this study integrates RT with the neighborhood relationship to construct a wide TWD framework. This framework incorporates decision-making references for the alternatives and accounts for the influence of experts’ psychological factors. Subsequently, the ranking of alternatives is determined using the technique for order preference by similarity to ideal solution (TOPSIS) method. Finally, the feasibility of the proposed method is demonstrated through a real-case study. Comparative experiments and parameter sensitivity analysis are designed to demonstrate the superiority and effectiveness of the model.}, archive = {J_ASOC}, author = {Zhanhao Liu and Huangjian Yi and Yushan Yao and Jiajia Wang}, doi = {10.1016/j.asoc.2025.113816}, journal = {Applied Soft Computing}, month = {12}, pages = {113816}, shortjournal = {Appl. Soft. Comput.}, title = {A regret theory-based three-way decision model under comparative linguistic expressions}, volume = {184}, year = {2025}, }
• (2025). A Q-learning based estimation of distribution algorithm for automated guided vehicle scheduling in disassembly workshop. ASOC, 184, 113815. (www)

  @article{ , abstract = {With the rapid upgrade of electronic products, the disassembly process is becoming increasingly crucial in facilitating resources recycling. In practical disassembly workshops, the automated guided vehicle (AGV) plays an important role in improving efficiency of product transportation between different disassembly machines. In this paper, a Q-learning based estimation of distribution algorithm (QEDA) is proposed to solve the AGV scheduling problem in disassembly workshops with minimization of makespan for all transportation tasks. Firstly, a mathematical model for the problem is established, and specific encoding and decoding rules are designed for solution representation. Secondly, a novel update mechanism of EDA is designed by fully information utilization of both elite and poor solutions to accelerate convergence. Thirdly, a local search strategy based on Q-learning with three time-related states is designed for the elite solutions to enhance exploitation. Finally, comparative experiments on 120 benchmark test sets demonstrate that the QEDA outperforms the state-of-the-art algorithms in both solution quality and convergence speed, confirming its superior effectiveness for AGV scheduling in disassembly workshops.}, archive = {J_ASOC}, author = {Honggui Han and Teng Wang and Jingjing Wang}, doi = {10.1016/j.asoc.2025.113815}, journal = {Applied Soft Computing}, month = {12}, pages = {113815}, shortjournal = {Appl. Soft. Comput.}, title = {A Q-learning based estimation of distribution algorithm for automated guided vehicle scheduling in disassembly workshop}, volume = {184}, year = {2025}, }
• (2025). Criterion-guided polypharmacy side effects prediction with dual-view contrastive learning. ASOC, 184, 113814. (www)

  @article{ , abstract = {Predicting side effects from taking multiple drugs (polypharmacy) is a critical challenge in healthcare, with significant implications for patient safety. Traditional machine learning methods often overlook the multi-scale nature of drug information and the interdependencies among side effects, limiting their predictive power. To address these gaps, we propose CASE (Criterion-guided polyphArmacy Side Effects prediction with dual-view contrastive learning). CASE captures both microscopic (molecular graph) and macroscopic (biochemical knowledge) drug features via an adaptive substructure encoder and a biochemical feature aggregator, respectively, under a dual-view framework. Then, contrastive learning is employed to jointly balance and integrate structural and biochemical representations, enabling a more comprehensive and synergistic understanding of drug interactions. Moreover, a criterion-guided decoding strategy is designed to model complex side effect relationships. This effective design enables the proposed CASE model to achieve superior performance, with an accuracy of 88.70 %, precision of 83.62 %, recall of 96.45 %, F1-score of 89.56 %, AUROC of 94.75 %, and AUPRC of 91.16 %, respectively, on the benchmark dataset. Ablation experiments and case studies further confirm the robustness and practical utility of the model. These results demonstrate that CASE can effectively assist clinical decision-making by identifying high-risk drug combinations and delivering reliable polypharmacy risk assessments, thereby supporting safer and more personalized treatment strategies.}, archive = {J_ASOC}, author = {Yike Wang and Huifang Ma and Zihao Gao and Zhixin Li and Liang Chang}, doi = {10.1016/j.asoc.2025.113814}, journal = {Applied Soft Computing}, month = {12}, pages = {113814}, shortjournal = {Appl. Soft. Comput.}, title = {Criterion-guided polypharmacy side effects prediction with dual-view contrastive learning}, volume = {184}, year = {2025}, }
• (2025). Identifying suicidal ideations from social media posts using deep learning and explainable AI-driven approach. ASOC, 184, 113813. (www)

  @article{ , abstract = {Background At present, suicide has become one of the leading causes of unnatural deaths worldwide. Individuals having suicidal urges often express their self-harming ideas through social media posts. Early identification of such ideations is critical for timely intervention and prevention. Besides, continuous assessment of the texts containing suicidal thoughts can uncover the hidden triggers of suicidal urges. This study presents a comprehensive approach to analyze the user-generated textual contents on social media that reflect suicidal ideas. Methodology For identifying the underlying topics that express suicidal ideations, this study has employed the Latent Dirichlet Allocation (LDA) model. Semantic Network Analysis (SNA) is used to gain a deeper quantitative and qualitative insight into these texts. Besides, an exploratory investigation of different deep learning (DL) models has been performed to identify the posts speculating suicidal ideations. Furthermore, this study has integrated Explainable AI (XAI) techniques like Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) to enhance interpretability of the decisions taken by the DL models. Techniques like LDA and SNA offer a better understanding of the linguistic features of the suicidal posts, while the integration of the XAI techniques with the DL models elevates the transparency of their decisions. Contributions This study has developed an end-to-end web application, that can perform real-time classification of posts for suicidal ideation. Moreover, this application can provide insights into the rationale behind the taken decisions. This study aims to contribute to suicide prevention efforts through an innovative combination of computational techniques and AI-driven tools.}, archive = {J_ASOC}, author = {Md. Sabab Zulfiker and Nasrin Kabir and Al Amin Biswas and Md. Mashih Ibn Yasin Adan and Mohammad Shorif Uddin}, doi = {10.1016/j.asoc.2025.113813}, journal = {Applied Soft Computing}, month = {12}, pages = {113813}, shortjournal = {Appl. Soft. Comput.}, title = {Identifying suicidal ideations from social media posts using deep learning and explainable AI-driven approach}, volume = {184}, year = {2025}, }
• (2025). A lightweight network enhanced by attention-guided cross-scale interaction for underwater object detection. ASOC, 184, 113811. (www)

  @article{ , abstract = {To address the image quality degradation caused by multipath effects and scattering in underwater environments, we propose a lightweight neural network architecture, PRCII-Net, optimized for small target detection under complex underwater conditions. First, a Progressive Re-parameterized Attention-based Intra-scale Feature Interaction module (PR-AIFI) is proposed, which improves the network performance while reducing the difficulty of training and maintaining training stability. Second, a feature pyramid network named as the Cross-scale Information Interaction Feature Pyramid Network (CII-FPN) is proposed, including the two main fusion structures. The CII-FPN not only fully utilizes shallow and deep information, but also enhances the network’s spatial representation and the interaction between deep feature layers, thereby boosting the detection capability for small targets. Meanwhile, to reduce the model’s size and resource consumption, 1x1 convolutions are introduced into the backbone network for efficient channel compression and cross-channel feature fusion, significantly lowering computational complexity. Experiments on the Detecting Underwater Objects (DUO) and Real-time Underwater Object Detection (RUOD) datasets demonstrate that PRCII-Net outperforms existing real-time neural network models in mAP and F1 scores while maintaining efficient inference on resource-constrained devices.}, archive = {J_ASOC}, author = {Dehua Zhang and Changcheng Yu and Zhen Li and Chunbin Qin and Ruixue Xia}, doi = {10.1016/j.asoc.2025.113811}, journal = {Applied Soft Computing}, month = {12}, pages = {113811}, shortjournal = {Appl. Soft. Comput.}, title = {A lightweight network enhanced by attention-guided cross-scale interaction for underwater object detection}, volume = {184}, year = {2025}, }
• (2025). Preferences relative ordering by frequency inclusion technique (PROFIT) for optimizing performance evaluation. ASOC, 184, 113807. (www)

  @article{ , abstract = {The aim of this investigation is to introduce the novel concept of the neutro-cardinal family of inclusion measures, a four-parameter family of fuzzy inclusion measures tailored for single-valued neutrosophic sets (SVNSs). These measures are built using a new scalar cardinality measure for SVNSs, termed weighted average cardinality, and incorporate logical operators such as single-valued neutrosophic Frank t-norms and their corresponding dual t-conorms. The versatility offered by this four-parameter family allows the variation of t-norms and t-conorms within each measure based on a fixed combination of parameters. This flexibly supports modeling various layers of human cognitive behavior, particularly the outermost states: optimism (via fuzzy Lukasiewicz t-norms and t-conorms), neutrality (via product and probabilistic sum operators), and pessimism (via min and max operators). Moreover, capturing these distinct mental states, the family allows combinations of them, enabling the representation of more complex and deeper cognitive processes. Consequently, the proposed inclusion measures serve as robust and versatile mathematical tools for designing artificial intelligence (AI) systems in multi-criteria decision making (MCDM) environments. Building upon this family, we present a new MCDM method named PROFIT (Preferences Relative Ordering by Frequency Inclusion Technique). PROFIT is designed as a simple yet effective decision-making approach that leverages the neutro-cardinal measures to support AI-based evaluation processes. It is investigated in the context of organizational management, in which it enables diligent performance evaluation. The PROFIT model’s data-driven and adaptive structure is consistent with resilience-focused initiatives like Resilient Manufacturing, ManuChain II, and Towards Resilience in Industry 5.0. As such, it serves as a strategic tool for optimizing decision-making, enhancing performance, and strengthening organizational resilience in dynamic and complex environments.}, archive = {J_ASOC}, author = {Madiha Qayyum and Dania Farooq and Muhammad Riaz and Muhammad Aslam and Dragan Pamucar}, doi = {10.1016/j.asoc.2025.113807}, journal = {Applied Soft Computing}, month = {12}, pages = {113807}, shortjournal = {Appl. Soft. Comput.}, title = {Preferences relative ordering by frequency inclusion technique (PROFIT) for optimizing performance evaluation}, volume = {184}, year = {2025}, }
• (2025). A state alignment-centric approach to federated system identification: The FedAlign framework. ASOC, 184, 113800. (www)

  @article{ , abstract = {This paper presents FedAlign, a Federated Learning (FL) framework, designed for System Identification (SYSID) of linear State-Space Models (SSMs) by aligning state representations. Local workers can learn linear SSMs with equivalent representations but different parameter basins. We demonstrate that directly aggregating these local SSMs via FedAvg results in a global model with altered system dynamics. FedAlign overcomes this problem by employing similarity transformation matrices to align state representations of local SSMs, thereby establishing a common parameter basin that retains the dynamics of local SSMs. FedAlign computes similarity transformation matrices via two distinct approaches. In FedAlign-A, we represent the global SSM in controllable canonical form (CCF). We use control theory to analytically derive similarity transformation matrices that convert each local SSM into this form. Yet, establishing global SSM in CCF brings additional alignment challenges in multi-input multi-output SYSID, as CCF representation is not unique, unlike in single-input single-output SYSID. In FedAlign-O, we address the alignment challenges by reformulating the local parameter basin alignment problem as an optimization task. We set the parameter basin of a local worker as the common parameter basin and solve least square problems to obtain the transformation matrices needed to align the remaining local SSMs. The experiments conducted on synthetic and real-world datasets show that FedAlign outperforms FedAvg, converges faster, and provides improved global SSM stability thanks to local parameter basins’ alignment.}, archive = {J_ASOC}, author = {Ertuğrul Keçeci and Müjde Güzelkaya and Tufan Kumbasar}, doi = {10.1016/j.asoc.2025.113800}, journal = {Applied Soft Computing}, month = {12}, pages = {113800}, shortjournal = {Appl. Soft. Comput.}, title = {A state alignment-centric approach to federated system identification: The FedAlign framework}, volume = {184}, year = {2025}, }
• (2025). Optimal harmonics prediction for distribution systems powered by multi-energy sources using bidirectional long-short term memory combined with data sequence. ASOC, 184, 113799. (www)

  @article{ , abstract = {A multi-energy resource aims to maintain a balance between energy output and load consumption and to ensure power continuity during different operating conditions. The harmonic distortions can be estimated from the output current of a harmonic source, which may not fully reflect its true harmonic distortions due to the interactions between the state changes at the power network level and the harmonic sources. System operators monitor each system's harmonic performance under different conditions of operation to find the actual contribution of grid-connected systems to harmonic-related issues. Development of machine learning algorithms leads to effective progress in the harmonic prediction and computation. In this paper, the combined data sequencing, and Bidirectional Long-Short Term Memory (Bi-LSTM) network has been exploited for the real-time harmonic prediction of future events in multi-energy sources. The validity of the proposed Model including the applications of ANFIS, ANNs, MLRA and LSTM is conducted on the two standard systems as IEEE 9-bus and IEEE 34-bus multi energy resources system that is associated with PV systems. The simulation results, based on climate changes of solar irradiance and ambient temperature in PV systems, demonstrate that the proposed methods can accurately forecast changes in total harmonic distortion (THD) as well as the voltage profile at the point of common coupling. The performance of Bi-LSTM, original LSTM, Machine Linear Regression (MLR), and Artificial Neural Networks (ANNs) techniques were assessed. These findings provide valuable insights. Four performance validation indices, RMSE, R-squared and MSE are considered to assess the performance of the competitive learning algorithms. The results showed that in the model IEEE 9-bus, Bi-LSTM outperformed all the applied methods as its RMSE value was 0.000019 while its MSE value was 3.61e-10 and finally, the Bi-LSTM had a higher value squared error (R 2 ) was equal 1 which indicates the effectiveness of Bi-LSTM for predicting sequential total harmonic distortion. On the other hand, in case study of IEEE 34-bus, the RMSE, MSE and R 2 are 0, 3.276e-30 and 1 using Bi-LSTM which means that the Bi-LSTM leads to the best performance validation indices compared to other competitive algorithms for the tested multi-energy systems.}, archive = {J_ASOC}, author = {Hasnaa M. El-Arwash and Almoataz Y. Abdelaziz and Ragab A. El-Sehiemy}, doi = {10.1016/j.asoc.2025.113799}, journal = {Applied Soft Computing}, month = {12}, pages = {113799}, shortjournal = {Appl. Soft. Comput.}, title = {Optimal harmonics prediction for distribution systems powered by multi-energy sources using bidirectional long-short term memory combined with data sequence}, volume = {184}, year = {2025}, }
• (2025). Advanced robot path planning on rough terrain: A Q-learning-based multi-objective PSO algorithm. ASOC, 184, 113798. (www)

  @article{ , abstract = {Path planning remains one of the most extensively studied problems in mobile robotics. While particle swarm optimization (PSO) has been widely adopted for this task, conventional implementations exhibit critical limitations including being prone to local optima, lacking population diversity, and having low accuracy — all of which compromise both the quality and efficiency of path planning solutions. To address these challenges, this paper proposes a novel hybrid algorithm called MOQLCPSO that synergistically integrates Q-learning (QL) with crossover operators into multi-objective particle swarm optimization (MOPSO) for car-like mobile robots operating in known static rough terrain environments. The proposed framework aims to generate collision-free optimal paths characterized by minimal length and terrain roughness through three key innovations: Firstly, we implement QL-based dynamic parameter adaptation to autonomously adjust PSO’s inertia weight and learning factors during optimization, effectively enhancing convergence towards the Pareto front. Secondly, a strategic crossover operator is introduced to augment exploration capabilities and maintain population diversity throughout the optimization process. Finally, comprehensive comparative simulations against state-of-the-art alternatives demonstrate MOQLCPSO’s superior performance metrics. The experimental results reveal statistically significant improvements in search accuracy, population diversity, and optimization efficiency across multiple terrain complexity levels. Notably, when benchmarked against the most competitive contemporary algorithm, MOQLCPSO achieves path length reduction of 0.64%, 2.42%, and 3.87%, terrain roughness reduction of 6.68%, 5.24%, and 4.72% in simple, moderately complex, and highly complex terrains, respectively. These advancements highlight the algorithm’s strong potential for deployment in multi-objective optimization scenarios ranging from autonomous mobile robot navigation and smart manufacturing resource allocation to power grid dispatch operations.}, archive = {J_ASOC}, author = {Zhaoxia Duan and Yi Zhang and Zhen Shao and Zhen Xu and Zhengrong Xiang}, doi = {10.1016/j.asoc.2025.113798}, journal = {Applied Soft Computing}, month = {12}, pages = {113798}, shortjournal = {Appl. Soft. Comput.}, title = {Advanced robot path planning on rough terrain: A Q-learning-based multi-objective PSO algorithm}, volume = {184}, year = {2025}, }
• (2025). Combined variable precision fuzzy rough set and its application in medical diagnosis. ASOC, 184, 113797. (www)

  @article{ , abstract = {In data processing, variable precision fuzzy rough set plays a crucial role in removing noisy data. Different variable precision ideas have been used to solve various problems and achieve different goals. Among them, the variable precision ideas by Zhao and Yao are widely recognized by researchers. However, a single method is often insufficient to solve complex problems. In this study, we combined these two variable precision ideas and introduced a new variable precision fuzzy rough set model ( O - C-VPFR ), which includes the advantages and properties of the aforementioned two models. Next, using the concept of attribute importance in O - C-VPFR , we developed a simple, objective method for calculating attribute weights. Based on this method, we further developed a comprehensive decision-making method integrating Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) and Preference Ranking Organization Method for Enrichment Evaluations (PROMETHEE). The new method was applied to the problem of health risk assessment of pregnant women (HRPW, https://archive.ics.uci.edu/dataset/863/maternal+health+risk ). We conducted parameter and comparative analyses by randomly selecting 10 data points from HRPW, proving the stability and reliability of our method through the Pearson correlation coefficient. Further, we used the entire dataset and performed an ordered similarity experiment and a hypothesis testing experiment to verify the stability, effectiveness, and robustness of the proposed method. In all experiments, our method comprehensively ranked pregnant women, identified high-risk individuals, and enabled timely treatment.}, archive = {J_ASOC}, author = {Jingwen Xie and Lingqiang Li}, doi = {10.1016/j.asoc.2025.113797}, journal = {Applied Soft Computing}, month = {12}, pages = {113797}, shortjournal = {Appl. Soft. Comput.}, title = {Combined variable precision fuzzy rough set and its application in medical diagnosis}, volume = {184}, year = {2025}, }
• (2025). Optimizing evacuation path planning in high-rise building fires using an improved ant colony algorithm and dynamic window approach under smoke control scenarios. ASOC, 184, 113796. (www)

  @article{ , abstract = {Traditional emergency evacuation strategies are constrained by limited adaptability to dynamic conditions and inefficient allocation of rescue resources. To assess the impact of smoke control facilities during high-rise fire evacuations, Building EXODUS software was used to simulate occupant behavior under the influence of air curtains and mechanical smoke exhaust systems. Furthermore, to address real-time obstacle avoidance in dynamic environments and during group movement in fire emergencies, a multi-person evacuation path planning method was developed by integrating an improved ant colony optimization algorithm with the dynamic window approach. This method incorporates a heuristic function that considers fire conditions and potential fields, along with a dynamic pheromone update strategy featuring reward and punishment mechanisms. The results indicate that the effectiveness of smoke control facilities in improving evacuation outcomes ranks as follows: combined operation of mechanical smoke extraction and air curtains, mechanical smoke extraction alone, and air curtains alone. Compared to the standard ant colony algorithm, the improved algorithm reduced planned path length by 7.71 %, decreased turn times by 73.30 %, and improved computational efficiency by 6.56 %. Furthermore, under scenarios with and without operational smoke control facilities, the improved algorithm combined with the dynamic window approach reduced path overlap areas by 11.76 % and 18.19 %, respectively, and shortened path lengths by 13.49 % and 14.19 %, respectively, compared to the Building EXODUS. The proposed method effectively addresses dynamic fire environments in high-rise buildings and provides a theoretical foundation for intelligent occupant evacuation.}, archive = {J_ASOC}, author = {Yaping Yu and Qinghe Wang and Lu Jin and Ji-nan Ding and Faqi Liu}, doi = {10.1016/j.asoc.2025.113796}, journal = {Applied Soft Computing}, month = {12}, pages = {113796}, shortjournal = {Appl. Soft. Comput.}, title = {Optimizing evacuation path planning in high-rise building fires using an improved ant colony algorithm and dynamic window approach under smoke control scenarios}, volume = {184}, year = {2025}, }
• (2025). Ribonucleic-acid protein interaction prediction based on deep learning: A comprehensive survey. ASOC, 184, 113795. (www)

  @article{ , abstract = {The interaction between Ribonucleic Acids (RNAs) and proteins, also called RNA Protein Interaction (RPI), governs biological processes, including gene regulation and disease pathogenesis. This comprehensive survey examines Artificial Intelligence (AI) applications in Deep Learning-based RPI Prediction (DL-based RPIP) through eight Research Questions (RQs), analyzing 179 studies (2014–2023). The key findings include: sustained technical evolution through embryonic (2014–2017), accelerated (2018–2022), and expansion phases (2023) (RQ1); hybrid models integrating Graph Neural Networks (GNNs) (for topological interface modeling) and Transformers (for long-range dependencies) achieve state-of-the-art performance (RQ4); pretrained language models enhance small-sample learning, but the cross-species generalization declines sharply with evolutionary distance (RQ5). Critical challenges persist, including data heterogeneity across databases, the scarcity of standardized benchmarks (RQ2), and balancing the trade-off between feature encoding and information preservation (RQ3). Future advancements require biologically informed DL architectures, multi-feature fusion, and rigorous cross-validation to bridge the generalization-interpretability gap (RQ8): This would accelerate the clinical translation of predictive tools (RQ6/RQ7). As the first comprehensive analysis spanning feature encoding, modeling, evaluation, applications, and tools, this work fills a critical gap in the DL-based RPIP literature.}, archive = {J_ASOC}, author = {Danyu Li and Rubing Huang and Chenhui Cui and Dave Towey and Ling Zhou and Jinyu Tian and Bin Zou}, doi = {10.1016/j.asoc.2025.113795}, journal = {Applied Soft Computing}, month = {12}, pages = {113795}, shortjournal = {Appl. Soft. Comput.}, title = {Ribonucleic-acid protein interaction prediction based on deep learning: A comprehensive survey}, volume = {184}, year = {2025}, }
• (2025). Multi-level pooling self-adaptive evolutionary multi-objective gene selection algorithm for microarray data classification. ASOC, 184, 113794. (www)

  @article{ , abstract = {Gene selection is a vital preprocessing technique for cancer classification using microarray data, focusing on identifying a subset of genes that achieve high classification accuracy while minimizing the gene selection rate. Despite the prevalence of multi-objective optimization algorithms in microarray data classification, balancing high classification accuracy and effective guidance in microarray classification remains challenging for most existing algorithms. This study proposes a novel multi-objective gene selection method, MOGS-MLPSAE, which utilizes multi-level pooling and self-adaptive evolutionary techniques to enhance classification accuracy. MOGS-MLPSAE employs a Pareto-based ranking pool division strategy to facilitate cross-level learning among individuals and introduces a population-biased evolutionary mechanism with five rules to drive the population toward higher classification accuracy. Compared with seven state-of-the-art multi-objective algorithms across 14 microarray datasets, MOGS-MLPSAE achieves superior performance, with classification accuracy 1.56–8.04 % higher than other algorithms and the lowest gene selection rate (average 1 %, minimum 0.01 %). This study demonstrates MOGS-MLPSAE's effectiveness in balancing classification accuracy and gene selection rate, offering a robust solution for microarray-based cancer classification.}, archive = {J_ASOC}, author = {Min Li and Rutun Cao and Chen Jin and Junke Wang and Shaobo Deng and Xiang Yu}, doi = {10.1016/j.asoc.2025.113794}, journal = {Applied Soft Computing}, month = {12}, pages = {113794}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-level pooling self-adaptive evolutionary multi-objective gene selection algorithm for microarray data classification}, volume = {184}, year = {2025}, }
• (2025). Activity transitions for semi-supervised federated learning in sensor-based human activity recognition. ASOC, 184, 113793. (www)

  @article{ , abstract = {Sensor-based Human Activity Recognition (HAR) is a core component in many real-world applications such as healthcare, fitness tracking, and smart environments. However, training effective HAR models is often constrained by limited labeled data and growing privacy concerns. Federated Learning (FL) offers a privacy-preserving solution by enabling collaborative model training without sharing raw sensor data. To further address the label scarcity challenge, we propose ATCoFed, a Semi-Supervised Federated Learning (SSFL) framework that introduces a novel pseudo-label filtering method based on activity transition patterns. Unlike existing approaches that rely solely on confidence thresholds, ATCoFed incorporates temporal context by using Long Short-Term Memory (LSTM) networks and Large Language Models (LLMs) as data-driven evaluators to validate the consistency of predicted activity sequences. This dual-evaluator mechanism improves the quality of pseudo-labels and enhances model robustness. Experimental results on benchmark HAR datasets demonstrate that ATCoFed consistently outperforms SSFL baselines, achieving better accuracy while maintaining computational and communication efficiency. These findings highlight the potential of activity-aware filtering to improve semi-supervised learning in privacy-preserving HAR applications.}, archive = {J_ASOC}, author = {Tori Andika Bukit and Ericka Pamela Bermudez Pillado and Bernardo Nugroho Yahya and Seok-Lyong Lee}, doi = {10.1016/j.asoc.2025.113793}, journal = {Applied Soft Computing}, month = {12}, pages = {113793}, shortjournal = {Appl. Soft. Comput.}, title = {Activity transitions for semi-supervised federated learning in sensor-based human activity recognition}, volume = {184}, year = {2025}, }
• (2025). TPCMEA: A tri-population evolutionary algorithm with adaptive stage-switching for complex CMOPs. ASOC, 184, 113792. (www)

  @article{ , abstract = {Constrained multi-objective optimization problems (CMOPs) involve the optimization of objective functions along with the satisfaction of constraints, presenting significant challenges in obtaining optimal solutions. To address difficulties such as conflicting objectives, complex constraints, disconnected and narrow feasible regions, this paper proposes a novel tri-population constrained multi-objective evolutionary algorithm (TPCMEA). This algorithm innovatively introduces an adaptive diversity-convergence stage-switching method (DCSSM), based on dual indicators of diversity and convergence, by dynamically monitoring the Euclidean distance distribution of the populations and convergence towards the optimal objective vector, overcoming the issues of premature convergence or resource wastage in traditional algorithms. Moreover, TPCMEA adopts a collaborative framework with three populations, each utilizing different search strategies to effectively solve complex CMOPs. These strategies encompass achieving rapid convergence under constraints, exploring unconstrained fronts to enhance solution diversity, and integrating information from the first two populations through knowledge transfer and compressive sampling strategies. Experimental results demonstrate that TPCMEA exhibits significant competitive advantages across the MW, LIRCMOP, CF benchmark test suites and real-world engineering problems, not only in terms of superior convergence and diversity but also displaying stronger robustness and adaptability in addressing issues related to narrow feasible regions and disconnected fronts.}, archive = {J_ASOC}, author = {Jun Chen and Xiaobo Li and Yuxin Zhao and Zhendi Ma and Zhongmei Han and Yanxia Bao}, doi = {10.1016/j.asoc.2025.113792}, journal = {Applied Soft Computing}, month = {12}, pages = {113792}, shortjournal = {Appl. Soft. Comput.}, title = {TPCMEA: A tri-population evolutionary algorithm with adaptive stage-switching for complex CMOPs}, volume = {184}, year = {2025}, }
• (2025). Multi-objective differential evolution algorithm integrating a directional generation mechanism for multi-objective optimization problems. ASOC, 184, 113791. (www)

  @article{ , abstract = {Multi-objective Evolutionary Algorithms (MOEAs) have gained significant attention due to their effectiveness in solving multi-objective optimization problems. However, when dealing with complex problems, they often face challenges such as low convergence accuracy and poor diversity. To address these issues, we propose a novel multi-objective differential evolution algorithm, MODE-FDGM, which integrates a directional generation mechanism. The key contributions are: (1) A directional-generation method leverages current and past information to rapidly build feasible solutions, boosting both speed and quality in exploring Pareto non-dominated space; (2) An update mechanism that combines crowding distance evaluation, iterating the population and incorporating historical information to enhance diversity and improve the ability to escape local optima; and (3) The introduction of an ecological niche radius concept along with a dual-mutation ecological niche selection evolution strategy, which improves exploration of unexplored spaces and preserves population diversity. Comparative experiments against 7 algorithms, including both classical and contemporary ones, on 24 benchmark functions demonstrate that the proposed algorithm markedly enhances the exploration of Pareto non-dominated solutions, exhibiting superior performance and advanced capabilities.}, archive = {J_ASOC}, author = {Zhuoxuan Yuan and Haibin Ouyang and Steven Li and Essam H. Houssein and Nagwan Abdel Samee}, doi = {10.1016/j.asoc.2025.113791}, journal = {Applied Soft Computing}, month = {12}, pages = {113791}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-objective differential evolution algorithm integrating a directional generation mechanism for multi-objective optimization problems}, volume = {184}, year = {2025}, }
• (2025). Relaxation-based exploration and clustering-based exploitation for multimodal multi-objective optimization. ASOC, 184, 113790. (www)

  @article{ , abstract = {Multimodal multi-objective optimization problems represent a specific type of multi-objective optimization problem where multiple distinct Pareto optimal solution sets (PSs) correspond to the same Pareto optimal front. Although the local PSs may not be as effective as the global PSs, obtaining the latter can be more costly. Therefore, the local PSs still hold some value for decision makers. However, many multimodal multi-objective evolutionary algorithms (MMOEAs) adopt the convergence-first criterion, which makes it challenging to obtain both the global and local PSs. To address the above issue, this paper proposes a two-stage dual-population coevolutionary algorithm with relaxation-based exploration and clustering-based exploitation (RCEA). Specifically, the relaxation-based exploration strategy defines an adaptive maximum Pareto rank, and relaxes the selection threshold for candidates from strictly non-dominated solutions to valuable solutions within this rank. Moreover, the clustering-based exploitation strategy performs a density-based decomposition on the population, separating different global and local PSs to facilitate more targeted evolution. During exploitation, a novel offspring generation operator is employed for intra-subpopulation crossover, preventing the waste of computational resources. Additionally, an improved convergence indicator is proposed to measure the convergence of solutions, which effectively handles the isolated solutions that may be erroneously identified as local Pareto optimal solutions by adaptively adjusting the neighborhood radius. Experimental results demonstrate that RCEA outperforms seven state-of-the-art MMOEAs on 70 benchmark problems.}, archive = {J_ASOC}, author = {Qianlong Dang and Zhiyang Zhang and Xiaochuan Gao and Tingting Wang}, doi = {10.1016/j.asoc.2025.113790}, journal = {Applied Soft Computing}, month = {12}, pages = {113790}, shortjournal = {Appl. Soft. Comput.}, title = {Relaxation-based exploration and clustering-based exploitation for multimodal multi-objective optimization}, volume = {184}, year = {2025}, }
• (2025). Compute-efficient and backpropagation-free pseudoinverse learning for neural networks: A comprehensive survey. ASOC, 184, 113789. (www)

  @article{ , abstract = {The pseudoinverse learning algorithm is a non-gradient, efficient learning scheme originally designed for training single hidden layer feedforward neural networks. It has been developed into various variants and successfully applied across numerous fields. This paper provides a systematic review of the fundamental theories of the pseudoinverse learning algorithm and its major variants, outlining different types of neural networks and learning system architectures based on the pseudoinverse learning scheme. Furthermore, we summarize the fundamental ideas and methodologies of applying the pseudoinverse learning scheme to various learning tasks such as classification, representation learning, time series forecasting, incremental learning, automated machine learning, and content generation. We also summarize and compare the performance of pseudoinverse learning with representative competing baselines on several commonly used data sets based on existing literature reports. The results demonstrate that PIL exhibits significant efficiency advantages over gradient-based approaches (training time was reduced by 72.73% to 99.37%), aligning with its inherent gradient-free nature. Notably, recent PIL variants maintain this computational superiority while achieving enhanced performance compared to other gradient-free algorithms. In addition, we briefly introduce the representative applications of pseudoinverse learning in various fields. To the best of our knowledge, this is the first comprehensive review in this field to encompass all aforementioned aspects. It facilitates the synthesis and integration of existing knowledge from disparate studies. By highlighting limitations in prior works including the computational complexity in large-scale pseudoinverse computation, potential numerical instability for ill-conditioned matrices, risk of overfitting, and constraints in modeling multidimensional patterns, this paper also recommends directions for future research in this area.}, archive = {J_ASOC}, author = {Ke Wang and Pandi Liu and Mohammed A.B. Mahmoud and Ping Guo and Yafei Li}, doi = {10.1016/j.asoc.2025.113789}, journal = {Applied Soft Computing}, month = {12}, pages = {113789}, shortjournal = {Appl. Soft. Comput.}, title = {Compute-efficient and backpropagation-free pseudoinverse learning for neural networks: A comprehensive survey}, volume = {184}, year = {2025}, }
• (2025). An efficient constrained multi-population cooperation search algorithm for multi-objective optimization of multireservoir operation. ASOC, 184, 113788. (www)

  @article{ , abstract = {In multi-objective optimization of multireservoir operation, complex nonlinear constraints often result in highly fragmented and discontinuous feasible regions, making it difficult for traditional algorithms to simultaneously achieve rapid convergence and maintain solution diversity. To effectively address this challenge, this paper proposes a novel Constrained Multi-Population Cooperation Search Algorithm (cMPCSA), which establishes an adaptive balance between global exploration and local exploitation through a cooperative evolutionary mechanism. The proposed algorithm achieves key advancements by constructing a modular multi-population framework with distinct subpopulation roles, integrating a constraint-independent environmental selection mechanism, and introducing a multi-factor learning strategy that combines edge, mean, and stochastic solutions to guide the search process more effectively. Experimental results on four mainstream multi-objective benchmark test suites demonstrate that cMPCSA significantly outperforms existing algorithms in terms of convergence performance and solution quality. Further application to a real-world multireservoir operation case confirms its robustness and superiority in coordinating multiple objectives such as hydropower generation, water supply reliability, and ecological flow protection. Overall, this study provides an efficient and scalable algorithmic tool for addressing complex multi-objective optimization problems in modern water resource management.}, archive = {J_ASOC}, author = {Zhong-kai Feng and Li Zhang and Xia-yu Wang and Fang Yang and Yi-hong Jiang and Sen Wang and Wen-jing Niu}, doi = {10.1016/j.asoc.2025.113788}, journal = {Applied Soft Computing}, month = {12}, pages = {113788}, shortjournal = {Appl. Soft. Comput.}, title = {An efficient constrained multi-population cooperation search algorithm for multi-objective optimization of multireservoir operation}, volume = {184}, year = {2025}, }
• (2025). Deep insights into cognitive decline: A survey of leveraging non-intrusive modalities with deep learning techniques. ASOC, 184, 113787. (www)

  @article{ , abstract = {Cognitive decline is a natural part of aging. However, under some circumstances, this decline is more pronounced than expected, typically due to disorders such as Alzheimer’s disease. Early detection of an anomalous decline is crucial, as it can facilitate timely professional intervention. While medical data can help, it often involves invasive procedures. An alternative approach is to employ non-intrusive techniques such as speech or handwriting analysis, which do not disturb daily activities. This survey reviews the most relevant non-intrusive methodologies that use deep learning techniques to automate the cognitive decline detection task, including audio, text, and visual processing. We discuss the key features and advantages of each modality and methodology, including state-of-the-art approaches like Transformer architecture and foundation models. In addition, we present studies that integrate different modalities to develop multimodal models. We also highlight the most significant datasets and the quantitative results from studies using these resources. From this review, several conclusions emerge. In most cases, text-based approaches consistently outperform other modalities. Furthermore, combining various approaches from individual modalities into a multimodal model consistently enhances performance across nearly all scenarios.}, archive = {J_ASOC}, author = {David Ortiz-Perez and Manuel Benavent-Lledo and Jose Garcia-Rodriguez and David Tomás and M. Flores Vizcaya-Moreno}, doi = {10.1016/j.asoc.2025.113787}, journal = {Applied Soft Computing}, month = {12}, pages = {113787}, shortjournal = {Appl. Soft. Comput.}, title = {Deep insights into cognitive decline: A survey of leveraging non-intrusive modalities with deep learning techniques}, volume = {184}, year = {2025}, }
• (2025). Improving adversarial transferability with neighborhood gradient information. ASOC, 184, 113786. (www)

  @article{ , abstract = {Deep neural networks (DNNs) are known to be susceptible to adversarial examples, leading to significant performance degradation. In black-box attack scenarios, a considerable attack performance gap between the surrogate model and the target model persists. This work focuses on enhancing the transferability of adversarial examples to narrow this performance gap. We observe that the gradient information around the clean image, i.e., Neighborhood Gradient Information (NGI) , can offer high transferability. Based on this insight, we introduce NGI-Attack, incorporating Example Backtracking and Multiplex Mask strategies to exploit this gradient information and enhance transferability. Specifically, we first adopt Example Backtracking to accumulate Neighborhood Gradient Information as the initial momentum term. Then, we utilize Multiplex Mask to form a multi-way attack strategy that forces the network to focus on non-discriminative regions, which can obtain richer gradient information during only a few iterations. Extensive experiments demonstrate that our approach significantly enhances adversarial transferability. Especially, when attacking numerous defense models, we achieve an average attack success rate of 95.2%. Notably, our method can seamlessly integrate with any off-the-shelf algorithm, enhancing their attack performance without incurring extra time costs.}, archive = {J_ASOC}, author = {Haijing Guo and Jiafeng Wang and Zhaoyu Chen and Kaixun Jiang and Lingyi Hong and Pinxue Guo and Jinglun Li and Wenqiang Zhang}, doi = {10.1016/j.asoc.2025.113786}, journal = {Applied Soft Computing}, month = {12}, pages = {113786}, shortjournal = {Appl. Soft. Comput.}, title = {Improving adversarial transferability with neighborhood gradient information}, volume = {184}, year = {2025}, }
• (2025). Federated semi-supervised learning for in-domain/Cross-domain person re-identification. ASOC, 184, 113785. (www)

  @article{ , abstract = {The Person Re-identification (Re-ID) task refers to retrieving images of the same person from different camera views. However, the task faces three major challenges: building a person dataset may infringe on personal privacy, collecting large-scale datasets for identity annotation requires high costs, and low quality local datasets can lead to performance bottlenecks. To address the above challenges, we propose Federated semi-supervised frameworks for the first time to solve in-domain/Cross-domain Re-ID problems, which can reduce annotation costs while protecting privacy. Firstly, we propose two different weighted aggregation federated learning strategies. Specifically, under the in-domain setting, labeled clients are utilized to determine the quality of unlabeled clients. Under the cross-domain setting, the model’s recognition capability is extended to each unlabeled domain by measuring inter-domain differences. In addition, a cascaded Global-personalized Model System and Personalized Model Knowledge Transfer Module are proposed to further address the domain-gap issue by sufficiently decoupling the global shared knowledge and the personalized knowledge unique to each domain. Finally, a Classifier Checking Module is proposed to significantly reduce computational costs by dynamically adjusting the pseudo-labels of unlabeled samples. Extensive experiments on multiple public datasets of varying scales and quality verify that the proposed frameworks can achieve acceptable performance at a lower cost while ensuring the security of personal data. Meanwhile, the effectiveness of the proposed training strategies and modules in other tasks is also verified, which proves their inspirational significance for research and applications in the fields of weakly supervised learning, user privacy protection and intelligent video surveillance.}, archive = {J_ASOC}, author = {Xinyuan Chen and Mingwen Shao and Yi Niu and Qiao Zhang}, doi = {10.1016/j.asoc.2025.113785}, journal = {Applied Soft Computing}, month = {12}, pages = {113785}, shortjournal = {Appl. Soft. Comput.}, title = {Federated semi-supervised learning for in-domain/Cross-domain person re-identification}, volume = {184}, year = {2025}, }
• (2025). Attention-guided triplet framework for synthetic data generation and semantic segmentation of railway fasteners under data scarcity and disparity. ASOC, 184, 113784. (www)

  @article{ , abstract = {Railway infrastructure monitoring faces significant challenges due to limited data availability, labor-intensive annotations, and imbalanced datasets—factors that hinder efficient fastener maintenance. To address these issues, this study proposes a novel three-phase framework that leverages attention-guided deep learning for the generation and analysis of synthetic railway fastener images. In the first phase, an attention-based Deep Convolutional Generative Adversarial Network (DCGAN) is introduced to generate high-fidelity synthetic images that closely mimic real-world conditions. Unlike conventional GANs, the attention mechanism enables the model to focus on critical structural features of fasteners, enhancing the realism and diversity of the generated data. The second phase applies advanced denoising techniques, with the DnCNN model outperforming traditional methods like Median Filtering in preserving fine details. The final phase employs a Convolutional Autoencoder (CAE) for accurate semantic segmentation, achieving 88.8 % accuracy on the synthetic dataset. This end-to-end methodology improves model generalizability, reduces reliance on manual labeling, and provides a cost-effective solution for automated railway inspection. By bridging the gap between real and synthetic data, it also lays the groundwork for scalable, intelligent infrastructure monitoring systems, supporting the advancement of safer and more efficient railway operations.}, archive = {J_ASOC}, author = {Qasim Zaheer and Momina Malik and S.Muhammad Ahmed Hassan Shah and Chengbo Ai and Hongzhi Wang and Zhiyu Liang and Shi Qiu}, doi = {10.1016/j.asoc.2025.113784}, journal = {Applied Soft Computing}, month = {12}, pages = {113784}, shortjournal = {Appl. Soft. Comput.}, title = {Attention-guided triplet framework for synthetic data generation and semantic segmentation of railway fasteners under data scarcity and disparity}, volume = {184}, year = {2025}, }
• (2025). An Actor–Critic-based adapted deep reinforcement learning model for multi-step traffic state prediction. ASOC, 184, 113783. (www)

  @article{ , abstract = {Traffic state prediction is critical to decision-making in various traffic management applications. Despite significant advancements in Deep Learning (DL) models, such as Long Short-Term Memory (LSTM), Graph Neural Networks (GNN), and attention-based transformer models, multi-step predictions remain challenging. The state-of-the-art models face a common limitation: the predictions’ accuracy decreases as the prediction horizon increases, a phenomenon known as error accumulation. In addition, with the arrival of non-recurrent events and external noise, the models fail to maintain good prediction accuracy. Deep Reinforcement Learning (DRL) has been widely applied to diverse tasks, including optimising intersection traffic signal control. However, its potential to address multi-step traffic prediction challenges remains underexplored. This study introduces an Actor–Critic-based adapted DRL method to explore the solution to the challenges associated with multi-step prediction. The Actor network makes predictions by capturing the temporal correlations of the data sequence, and the Critic network optimises the Actor by evaluating the prediction quality using Q-values. This novel combination of Supervised Learning and Reinforcement Learning (RL) paradigms, along with non-autoregressive modelling, helps the model to mitigate the error accumulation problem and increase its robustness to the arrival of non-recurrent events. It also introduces a Denoising Autoencoder to deal with external noise effectively. The proposed model was trained and evaluated on three benchmark traffic flow and speed datasets. Baseline multi-step prediction models were implemented for comparison based on performance metrics such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The results reveal that the proposed method outperforms the baselines by achieving average improvements of 0.26 to 21.29% in terms of MAE and RMSE for up to 24 time steps of prediction length on the three used datasets, at the expense of relatively higher computational costs. On top of that, this adapted DRL approach outperforms traditional DRL models, such as Deep Deterministic Policy Gradient (DDPG), in accuracy and computational efficiency.}, archive = {J_ASOC}, author = {Selim Reza and Marta Campos Ferreira and J.J.M. Machado and João Manuel R.S. Tavares}, doi = {10.1016/j.asoc.2025.113783}, journal = {Applied Soft Computing}, month = {12}, pages = {113783}, shortjournal = {Appl. Soft. Comput.}, title = {An Actor–Critic-based adapted deep reinforcement learning model for multi-step traffic state prediction}, volume = {184}, year = {2025}, }
• (2025). Grammatical evolution for automatic design of actuated traffic signal control plans. ASOC, 184, 113782. (www)

  @article{ , abstract = {In traffic networks, proper signal control design is essential to ensure a reasonable level of service. Signal control designs are becoming increasingly complex, with numerous settings that must be calibrated and set. This paper introduces a novel approach for handling this complexity by automatically generating optimal actuated signal control plans using Grammatical Evolution (GE). GE has proven its effectiveness in automating the design of different complex systems, such as neural networks and analog electronic circuits. GE’s distinctive mapping and representation capabilities make it a powerful candidate for optimizing various systems. In contrast to traditional optimization methods for actuated signal plans, which focus on specific parameters, such as green times and cycle length, the GE-based approach evolves complete plans, including phases, detector placements, and transit priority strategies. As a result, it eliminates the need for human intervention in the design process, making it more efficient and less time-consuming. The proposed approach was tested with an application to an isolated intersection in Haifa, Israel. The results showed that the automatically generated signal plan outperformed the existing plan by reducing delay times and queue lengths. Moreover, this method demonstrated its efficiency in generating reliable traffic signal plans under challenging traffic conditions.}, archive = {J_ASOC}, author = {Mahmud Keblawi and Tomer Toledo}, doi = {10.1016/j.asoc.2025.113782}, journal = {Applied Soft Computing}, month = {12}, pages = {113782}, shortjournal = {Appl. Soft. Comput.}, title = {Grammatical evolution for automatic design of actuated traffic signal control plans}, volume = {184}, year = {2025}, }
• (2025). Optimized resource management in containerized clouds via hierarchical autoregressive network based workload prediction. ASOC, 184, 113781. (www)

  @article{ , abstract = {The swift growth of containerized applications calls for highly precise and real-time prediction of workloads to make resource allocation in cloud environments efficient. Current methods are deficient either by not being able to catch the complex long-term workload dependencies in patterns or by introducing too much computational overhead that degrades real-time performance. We propose a new integrated model where a transformer-based Hierarchical Autoregressive Network (HARN) is dynamically fused with TES. With the use of self-attention mechanisms in modeling complex temporal dynamics and long-range dependencies, on the one hand, HARN captures the detailed seasonal variation but robustly deals with the smoother short-term oscillations, giving TES ample opportunity to function. The model dynamically adjusts their contributions to fit the best weighting in real-time, hence leading to optimal predictability even for changing workloads. Our analysis on the publicly released Alibaba v2018 dataset shows an average improvement in Mean Absolute Error (MAE) of 15.50% and an improvement in Root Mean Square Error (RMSE) of 14.80%. Furthermore, experiments on a specially designed container dataset show even more significant improvements, with improvements of up to 25% in CPU metrics and 40% in memory metrics. These findings and minimal computational overhead emphasize the model’s ability to facilitate real-time proactive resource provisioning in dynamic cloud settings.}, archive = {J_ASOC}, author = {Shivani Tripathi and Angelina Shibu and Priyadarshni and Rajiv Misra and T.N. Singh}, doi = {10.1016/j.asoc.2025.113781}, journal = {Applied Soft Computing}, month = {12}, pages = {113781}, shortjournal = {Appl. Soft. Comput.}, title = {Optimized resource management in containerized clouds via hierarchical autoregressive network based workload prediction}, volume = {184}, year = {2025}, }
• (2025). Pervasive multifaceted process based generative adversarial network for image quality enhancement. ASOC, 184, 113780. (www)

  @article{ , abstract = {The practice of Generative Adversarial Networks (GAN) has extended a lot of consideration in recent times. In most of the GAN methods are problem-specific related that are personalized to report numerous trials of individual application rather than performing other image improvement tasks. Furthermore, the basic GAN generators influence their boundaries in numerous image restoration and development use cases. Therefore, in this paper, we propose a generic GAN referred to as Pervasive Multifaceted Process based Generative Adversarial Network (PMPGAN). In this generator, we introduced multiple Convolutional Neural Networks (CNN) followed by Multi-dimensional Pyramid Pooling Module (MPPM) and Attention Module (AM), of which the input for the AM is given as low-level features and it produces output with enhanced feature map. Meanwhile, we enhanced the improvement outcome of the generated image with discriminator loss function. Finally, we tested the efficiency of the proposed system through extensive experiments on five challenging applications for image enhancement, image restoration, and infrared image translation to determine the dominance and efficiency in eliminating image degradation and producing visually interesting fake images. Our PMPGAN quantitatively outperforms several latest models. The results show that the PMPGAN model is superior to the existing models.}, archive = {J_ASOC}, author = {K. Balaji}, doi = {10.1016/j.asoc.2025.113780}, journal = {Applied Soft Computing}, month = {12}, pages = {113780}, shortjournal = {Appl. Soft. Comput.}, title = {Pervasive multifaceted process based generative adversarial network for image quality enhancement}, volume = {184}, year = {2025}, }
• (2025). Long-term time series forecasting by a frequency-domain enhanced temporal convolutional network with the stationary residual regularization. ASOC, 184, 113779. (www)

  @article{ , abstract = {The current time series models can resolve the long-term time series forecasting problem by leveraging artificial neural networks (ANNs) to capture complex temporal dependencies. However, the state-of-the-art ANNs for long-term prediction, Transformer-based models, have two inherent shortcomings that undermine both credibility and accuracy: losing temporal order information and ignoring information in prediction residuals. To fill these gaps, this study proposes a frequency-domain enhanced multi-scale temporal convolutional network (FMTCN) and a hybrid loss function regularized by stationary residuals called QM. Firstly, to preserve temporal order information, a multi-scale block with dilated causal convolution is developed as the core component of FMTCN, which can ensure the dependencies of predictions on sequentially ordered data. Secondly, an adaptive fusion mechanism for time and frequency domain patterns is designed to bridge the performance gap between convolution-based and Transformer-based models. Thirdly, to reduce residual autocorrelation, the proposed hybrid loss function integrates the Ljung–Box statistic into the mean squared error loss function as the regularization. Finally, extensive experiments across five real-world datasets are conducted to validate the proposed methods. Compared with the state-of-the-art method, the proposal improves the accuracy by 8.6% in the univariate long-term prediction and 6.1% in the multivariate long-term prediction.}, archive = {J_ASOC}, author = {Jing Wang and Yanbing Ju and Peiwu Dong and Tian Ju}, doi = {10.1016/j.asoc.2025.113779}, journal = {Applied Soft Computing}, month = {12}, pages = {113779}, shortjournal = {Appl. Soft. Comput.}, title = {Long-term time series forecasting by a frequency-domain enhanced temporal convolutional network with the stationary residual regularization}, volume = {184}, year = {2025}, }
• (2025). Loss functions in classification: An comprehensive overview and comparative study. ASOC, 184, 113778. (www)

  @article{ , abstract = {Loss functions are among the most fundamental components of a classifier’s learning processes and consequently can significantly affect the performance of classification methods. However, despite the substantial potential impact of loss functions on the classification accuracy, they have been incomprehensively investigated in the literature. For this reason, this paper tries to comprehensively evaluate the potency and impact of loss functions on the performance and accuracy in classification problems. For this purpose, fifty-five different loss functions in twelve categories—Linear Continuous Distance, Nonlinear Continuous Distance, Linear Semi-Continuous Distance, Nonlinear Semi-Continuous Distance, Linear Discrete Distance, Nonlinear Discrete Distance, Linear Continuous Direction, Nonlinear Continuous Direction, Linear Semi-Continuous Direction, Nonlinear Semi-Continuous Direction, Linear Discrete Direction, and Nonlinear Discrete Direction, are considered. In addition, in this paper, six distinct environmental pollutants/pollution benchmark data sets, three classifier types—statistical, shallow intelligence, and deep intelligence, are exemplary considered. Furthermore, in this paper, the most important measure of regular classification problems, the classification rate, has been chosen in order to compare these loss functions. The experimental results have unequivocally validated that the choice of loss functions has a significant impact on accuracy and classification rates. Numerical results of loss functions indicate that the difference between the lowest and the highest classification rate in the statistical, shallow intelligent, and deep learning classifiers, is averagely equal to 6.41 %, 5.16 %, and 2.24 %, respectively. It means that the model designer, by choosing the appropriate loss function, can averagely improve more than 4 % the obtained classification rate. Empirical results show that in the general perspective, the lowest performances of loss functions are overall related to the linear, continuous, and distance-based categories, in contrast to the highest ones, which are nonlinear, discrete, and direction-based. The exception to this outcome is the Zero-One family, in which all their loss functions, i.e., linear/nonlinear, distance/direction, are equivalent and yield the same accuracy. The best classification rate is also related to this family, that significantly better than other loss functions in all cases, as well as all classifiers. The Zero-One loss functions can averagely achieve a 98.62 % classification rate that 17.65 % is averagely higher than others. These evidences illustrate that, in addition to other effective features and factors, the loss function type should also be considered by environmental model designers in order to yield a more desired classification rate.}, archive = {J_ASOC}, author = {Fatemeh Chahkoutahi and Mehdi Khashei and Naser Molaverdi}, doi = {10.1016/j.asoc.2025.113778}, journal = {Applied Soft Computing}, month = {12}, pages = {113778}, shortjournal = {Appl. Soft. Comput.}, title = {Loss functions in classification: An comprehensive overview and comparative study}, volume = {184}, year = {2025}, }
• (2025). Ultra-low memory spatiotemporal decomposition recurrent neural networks for edge structural fault monitoring. ASOC, 184, 113777. (www)

  @article{ , abstract = {The hardware memory resources of microcontrollers in wireless sensor network nodes are limited, currently only capable of data acquisition and simple computing, making it difficult to perform neural network inference for edge fault monitoring. Running neural networks on microcontrollers can enhance the edge data processing capabilities of the nodes. Recurrent neural networks with short and medium term memory excel at processing sequential data. However, inference on microcontrollers consumes a significant amount of memory, necessitating solutions to the memory constraints. This study proposes a spatiotemporal decomposition method for recurrent neural networks to address the memory constraint issue when performing inference on resource-constrained nodes, thus enabling edge fault monitoring. The method decomposes recurrent neural networks in spatiotemporal dimensions, significantly reducing memory usage during operation. Additionally, it proposes model parameter storage and addressing techniques to ensure accurate reading of Flash data. Experiments have verified that this method can achieve 99.7 % accuracy in edge fault classification, consuming only 332 bytes of RAM and 768 bytes of Flash. The spatiotemporal decomposition method also provides a solution for running other time-series models on resource-constrained edge devices.}, archive = {J_ASOC}, author = {Hao Fu and Lei Deng and Baoping Tang and Shuaiwen Cui and Yuguang Fu}, doi = {10.1016/j.asoc.2025.113777}, journal = {Applied Soft Computing}, month = {12}, pages = {113777}, shortjournal = {Appl. Soft. Comput.}, title = {Ultra-low memory spatiotemporal decomposition recurrent neural networks for edge structural fault monitoring}, volume = {184}, year = {2025}, }
• (2025). Hybrid SpinalNet-fuzzy-shufflenet for brain tumor detection using MRI images. ASOC, 184, 113775. (www)

  @article{ , abstract = {Today, a significant portion of the global population is affected by the serious medical condition known as Brain Tumor (BT). Damage to healthy brain tissue is suspected, as it is currently the most important cause of a huge quantity of mortality. To prevent patients from dying, early detection is very essential. Despite various notable efforts and hopeful results in this area, accurate segmentation and categorization remain a difficult task. To address these gaps, a SpinalNet Fuzzy Shufflenet (SFShuffleNet) is proposed for the detection of BT. First, the input image is fed into the preprocessing stage, utilizing ROI extraction. Subsequently, the preprocessed image undergoes enhancement using histogram equalization techniques. Then, the enhanced image undergoes segmentation with Fuzzy Local Information C-Means (FLICM). Following segmentation, the image is augmented through sharpening, translation, random erasing, and resizing techniques. Features such as mean, variance, standard deviation, average, contrast, skewness, kurtosis, and entropy with the Local Frequency Descriptor (LFD) are extracted. Finally, the proposed SFShuffleNet, which combines SpinalNet and Shufflenet with fuzzy concept modifications, detects BT. SFShuffleNet achieved the highest accuracy, sensitivity, specificity, and F1-score of 91.46 %, 90.96 %, 92.78 %, and 90.79 %, respectively.}, archive = {J_ASOC}, author = {P Srinivasa Rao and Swathi Sowmya Bavirthi and G. Sharada and Ponnaboyina Ranganath and Vemuri Sailaja and G Vimala Kumari}, doi = {10.1016/j.asoc.2025.113775}, journal = {Applied Soft Computing}, month = {12}, pages = {113775}, shortjournal = {Appl. Soft. Comput.}, title = {Hybrid SpinalNet-fuzzy-shufflenet for brain tumor detection using MRI images}, volume = {184}, year = {2025}, }
• (2025). Enzyme classification integrating LSTM and prot-BERT sequence encoding. ASOC, 184, 113774. (www)

  @article{ , abstract = {Enzyme classification is essential for deciphering cellular and biological processes, driving targeted research, and influencing domains such as drug discovery and bioengineering. While various automated tools exist for enzyme classification, many are limited in scope or no longer operational. This study utilizes advanced artificial intelligence (AI) algorithms, including SVM, RF, simpleRNN, LSTM, ConvLSTM, and bidirectional LSTM, combined with numeric and Prot-BERT-based protein sequence encodings, to classify 1991 enzymes across 7 main classes, 69 subclasses, 216 sub-subclasses, and 1333 substrates. Among all trained models, the bidirectional LSTM model integrated with Prot-BERT sequence encoding (ECiLPSE) demonstrated exceptional accuracy of 99.14 % on the training set and 98.41 % on the test set, effectively capturing intricate sequence details. In addition to high accuracy, ECiLPSE achieved an F1-score of 0.99 (training set) and 0.98 (test set), with AU-ROC scores of 0.98 and 0.97, respectively. Precision and recall were both 0.99 on the training set and 0.98 on the test set. The 95 % confidence interval for test set accuracy (98.00 % - 99.97 %) further supports the model’s robustness and discriminative capability across classes. To benchmark performance, ECiLPSE was evaluated against four EC prediction tools (ECPred, CLEAN, EZYPred, EzyDeep), demonstrating superior accuracy and computational efficiency. The results from three case studies (using New-1277, Price-149, and halogenases datasets) highlight the limitations of the existing tools, underscoring the predictive accuracy, reliability, and applicability of ECiLPSE. ECiLPSE is available as a standalone tool for large datasets and as a web server for small datasets, offering a robust resource for enzyme classification, advancing computational approaches in enzyme research, and promising applications in drug discovery.}, archive = {J_ASOC}, author = {Anju Sharma and Vineet Diwakar and Rajnish Kumar and Prabha Garg}, doi = {10.1016/j.asoc.2025.113774}, journal = {Applied Soft Computing}, month = {12}, pages = {113774}, shortjournal = {Appl. Soft. Comput.}, title = {Enzyme classification integrating LSTM and prot-BERT sequence encoding}, volume = {184}, year = {2025}, }
• (2025). Intelligent urban GNSS measurement uncertainty prediction by exploring the spatial characteristics with transformer. ASOC, 184, 113773. (www)

  @article{ , abstract = {Global navigation satellite system (GNSS) positioning accuracy is essential for civil applications. Unfortunately, in urban areas, GNSS signals are easily blocked and reflected by tall buildings, namely non-line-of-sight (NLOS) receptions. Thus, it is essential to classify these contaminated receptions and mitigate their pseudorange error for positioning improvement. This paper designed a Transformer-based network that utilizes satellite spatial characteristics to predict GNSS measurement uncertainty. The proposed method accurately classifies 89 % of satellites’ visibility and around 45 % compensation in NLOS pseudorange error, outperforming the state-of-the-art algorithms. By analyzing the attention matrix generated by Transformer, we explore how Transformer utilizes spatial characteristics for satellite measurement uncertainty prediction, which regards satellite visibility and pseudorange as different tasks for training.}, archive = {J_ASOC}, author = {Zekun Zhang and Penghui Xu and Guohao Zhang}, doi = {10.1016/j.asoc.2025.113773}, journal = {Applied Soft Computing}, month = {12}, pages = {113773}, shortjournal = {Appl. Soft. Comput.}, title = {Intelligent urban GNSS measurement uncertainty prediction by exploring the spatial characteristics with transformer}, volume = {184}, year = {2025}, }
• (2025). Modeling self-supervised learning for handling the concept drift with ternary-adaptive ensemble. ASOC, 184, 113772. (www)

  @article{ , abstract = {Over the last several decades, drift detection and adaptation models have been trained on historical drift labels struggle with real-world data streams due to evolving distributions, impacting accurate future predictions. Existing methods update predictive models but often fail to retain previously learned patterns from streaming data, hindered by insufficient predefined labels for categorizing drift detection. To address these constraints, the proposed work introduces a two-task framework comprising a Self-Supervised Learning (SSL) model with a pretext task and a downstream task to efficiently adapt the model to evolving data distributions. In the pretext task, the proposed approach learns patterns from the unlabeled offline data in a self-supervised manner and utilizes the knowledge from the performance-based drift detector. In a subsequent downstream task, the model that has been enhanced with a forgetting-aware Ternary-Adaptive Ensemble (TAE) learner without requiring the drift labels. In the proposed approach, the TAE algorithm comprises drift-aware model updation methods for the Long Short-Term Memory (LSTM) to resolve forgetting in sequential data streams, and the three updation methods involve the pretext model’s weight transfer, Bayesian Optimization-based hyperparameter tuning, and Elastic Weight Consolidation (EWC). The proposed approach effectively handles drifts and outperformed while testing on the real-world weather and synthetic, mixed drift datasets with 82.6% and 92.36% accuracy, respectively.}, archive = {J_ASOC}, author = {Shubhangi Suryawanshi and Anurag Goswami and Pramod Patil}, doi = {10.1016/j.asoc.2025.113772}, journal = {Applied Soft Computing}, month = {12}, pages = {113772}, shortjournal = {Appl. Soft. Comput.}, title = {Modeling self-supervised learning for handling the concept drift with ternary-adaptive ensemble}, volume = {184}, year = {2025}, }
• (2025). Cross-domain knowledge distillation for domain adaptation with GCN-driven MLP generalization. ASOC, 184, 113771. (www)

  @article{ , abstract = {Knowledge distillation (KD) and domain adaptation (DA) represent potential research directions for reducing costs associated with deploying deep neural networks (DNN) in real-world applications. KD focuses on model compression, exploring methods to transfer informative representations from a complex model to a lighter one without incurring additional costs. Conversely, DA emphasizes the data distribution perspective, aiming to decrease labeling expenses by leveraging knowledge extracted from a labeled source domain to minimize classification errors in an unlabeled target domain. In this paper, we introduce a novel knowledge distillation (KD) approach with a teacher–student paradigm for domain adaptation (DA) tasks, termed Improved Cross-domain Knowledge Distillation (ICDKD). Specifically, we employ a graph convolutional network (GCN) classifier as the teacher model and a multilayer perceptron (MLP) classifier as the student model. During training, the teacher model utilizes a message-passing mechanism to capture the topology of the training data through neighbor information, thus explicitly enhancing semantic representations in each category to improve classification accuracy. Subsequently, the extracted knowledge from the GCN teacher model is distilled to the MLP student model. Finally, in the inference stage, only the MLP student model is utilized to meet the latency constraints of applications. Our proposed method effectively combines the strengths of both GCN and MLP classifiers to improve the classification performance and satisfy the real-world application requirements. We implemented our method on various DA benchmark datasets under unsupervised and semi-supervised domain adaptation settings, including ImageCLEF-DA, Office-31, Office-Home, VisDA2017, and DomainNet. The experimental results demonstrate the effectiveness of our proposed method on both CNN-based and ViT-based architectures, achieving outstanding classification performance compared to prior state-of-the-art domain adaptation methods.}, archive = {J_ASOC}, author = {Ba Hung Ngo and Tae Jong Choi}, doi = {10.1016/j.asoc.2025.113771}, journal = {Applied Soft Computing}, month = {12}, pages = {113771}, shortjournal = {Appl. Soft. Comput.}, title = {Cross-domain knowledge distillation for domain adaptation with GCN-driven MLP generalization}, volume = {184}, year = {2025}, }
• (2025). A fuzzy delphi-SERVQUAL model using degree of belief structure for assessing customer satisfaction in automotive after-sales services. ASOC, 184, 113770. (www)

  @article{ , abstract = {This study presents a novel two-phase fuzzy multi-attribute decision-making (MADM) framework for assessing customer satisfaction in automotive after-sales services under uncertainty. The proposed model integrates a modified Fuzzy Delphi Method (FDM) with an enhanced fuzzy SERVQUAL approach, both embedded within a Degree of Belief (DoB) structure to capture evaluative ambiguity and confidence levels in expert and customer judgments. In the first phase, a belief-based FDM enables the efficient screening and prioritization of 58 service quality criteria using a single-round process that incorporates both optimistic and pessimistic expert viewpoints. In the second phase, a belief-driven SERVQUAL model evaluates customer perceptions and expectations from both stringent and lenient perspectives across six dimensions, including a newly introduced Digital Technology dimension that reflects emerging service delivery mechanisms. The model also features a δ-based sensitivity analysis to examine the impact of decision-making attitudes and a seven-zone classification system to categorize service quality gaps with high diagnostic precision. Application of the proposed framework in the Iranian automotive after-sales sector enabled the prioritization of 29 key service quality criteria from an initial pool of 58 indicators, with the most critical factors identified in the ‘Reliability’ and ‘Responsiveness’ dimensions. The resulting insights support evidence-based managerial actions, including the reallocation of resources toward high-impact service areas, targeted digital transformation initiatives, and the formulation of differentiated improvement strategies based on gap severity and belief-based customer expectations .}, archive = {J_ASOC}, author = {Mojtaba Elahi and Ramin Enayati and Mehdi Keramatpour}, doi = {10.1016/j.asoc.2025.113770}, journal = {Applied Soft Computing}, month = {12}, pages = {113770}, shortjournal = {Appl. Soft. Comput.}, title = {A fuzzy delphi-SERVQUAL model using degree of belief structure for assessing customer satisfaction in automotive after-sales services}, volume = {184}, year = {2025}, }
• (2025). Developing a hybrid circular intuitionistic fuzzy framework to assess tourism 5.0 challenges. ASOC, 184, 113769. (www)

  @article{ , abstract = {The adoption of Industry 5.0 (I5.0) in tourism is still in its early stages, and its complex implementation presents challenges that are not fully understood. Existing studies often focus on context-specific challenges, limiting the exploration of broader obstacles. This research aims to identify a comprehensive set of challenges, rank them, and evaluate their interdependencies in I5.0 adoption within tourism. A systematic literature review was conducted to identify key challenges, and expert input was gathered to validate and refine the findings. To evaluate the significance of the identified challenges, a novel extension of the Stepwise Weight Assessment Ratio Analysis (SWARA) method into the Circular Intuitionistic Fuzzy (CIF) environment, referred to as CIF-SWARA, was formulated to determine the relative weights of each challenge under uncertainty with more precision. Additionally, the Decision-Making Trial and Evaluation Laboratory (DEMATEL) method was extended using CIF sets, termed CIF-DEMATEL, to evaluate the causal relationships and relative influence among the challenges. The results from both methods were combined to derive final weights. Furthermore, a conceptual model was constructed using Total Interpretive Structural Modeling (TISM) based on the CIF-DEMATEL output, providing a detailed qualitative assessment of the interactions among the challenges. Findings highlight that “infrastructure issues” are the most urgent challenge requiring attention. Additionally, “technical and technological issues,” “acceptance and adaptability issues,” and “financial issues” emerged as the three most significant challenges for transitioning to “Tourism 5.0.” The model’s robustness was confirmed via sensitivity analysis, and comparative evaluations demonstrated consistent performance against established fuzzy environments. This study advances I5.0 research in tourism by identifying and modeling key challenges, offering valuable insights for policymakers.}, archive = {J_ASOC}, author = {Vahideh Shahin and Dragan Pamucar and Moslem Alimohammadlou}, doi = {10.1016/j.asoc.2025.113769}, journal = {Applied Soft Computing}, month = {12}, pages = {113769}, shortjournal = {Appl. Soft. Comput.}, title = {Developing a hybrid circular intuitionistic fuzzy framework to assess tourism 5.0 challenges}, volume = {184}, year = {2025}, }
• (2025). A self-adaptive discrete artificial bee colony algorithm based on block swap for steelmaking and continuous casting scheduling problem. ASOC, 184, 113768. (www)

  @article{ , abstract = {Due to the high temperature characteristics of the steelmaking continuous casting production, the temperature decreasing energy consumption (TDEC) in the processing process cannot be ignored. Therefore, this paper establishes a mixed integer mathematical model for the steelmaking continuous casting scheduling problem with TDEC (SCCSP TDEC ). The SCCSP TDEC chooses to minimize the objective functions of the total weighted earliness and tardiness, the TDEC and the makespan. To solve the SCCSP TDEC , an improved heuristic based on problem-specific features is proposed in this paper, which takes into account the completion time of the third stage’s cast, and the effect of the cast’s holistic nature. Then, a self-adaptive discrete artificial bee colony algorithm based on block swap (SDABC bs ) is proposed to solve the SCCSP TDEC . This strategy can adaptively select an optimal structure suitable for the current population, and effectively expand the search space of neighborhood solutions. Finally, this paper tests the SDABC bs algorithm based on the extensive instances generated according to the actual production process, and the performance of the SDABC bs algorithm is verified by comparing with other efficient algorithms. The experimental results show that the proposed SDABC bs algorithm is more effective for the SCCSP TDEC among all the algorithms in comparison.}, archive = {J_ASOC}, author = {Yang Yu and Guo-Dong Yang and Qichun Zhang and Liangliang Sun and Xinfu Pang and Yefeng Liu}, doi = {10.1016/j.asoc.2025.113768}, journal = {Applied Soft Computing}, month = {12}, pages = {113768}, shortjournal = {Appl. Soft. Comput.}, title = {A self-adaptive discrete artificial bee colony algorithm based on block swap for steelmaking and continuous casting scheduling problem}, volume = {184}, year = {2025}, }
• (2025). Fine-grained mutation operators for community hiding using genetic algorithms. ASOC, 184, 113767. (www)

  @article{ , abstract = {Community detection plays a key role in uncovering group structures in networks, but its misuse can lead to privacy risks by exposing sensitive relationships. As a proactive defense, community hiding seeks to perturb the network structure to reduce the effectiveness of community detection algorithms. Given the NP-hard nature of this task, genetic algorithms (GAs) widely used due to their robust global search capabilities. However, existing methods lack effective guidance when dealing with vast solution spaces, resulting in inefficient exploration and suboptimal obfuscation outcomes. To address this, we propose Network- T opology- C ombined Community Information H iding A lgorithm (TCHA), a novel GA-based method that leverages node similarity information via node embedding to guide perturbations more effectively. TCHA introduces a multidimensional mutation operator that combines coarse-grained and fine-grained mutation strategies. These fine-grained mutations are performed in the embedding space and decoded back to graph edits, enabling more precise and topologically-aware perturbations. To evaluate the efficiency of this process, we introduce a novel metric based on expected path length within a mutation transition graph, offering deeper insight into evolutionary search dynamics. Experiments on six real-world networks demonstrate that TCHA achieves an average modularity reduction of 31.92% and an average normalized mutual information of 0.6447, outperforming baselines such as Q-Attack, NEURAL, and DICE. These results confirm the superiority of the embedding-guided fine-grained mutation strategy in enhancing community hiding effectiveness.}, archive = {J_ASOC}, author = {Shanqing Yu and Jintao Zhou and Meng Zhou and Yidan Song and Jiaxiang Li and Zeyu Wang and Qi Xuan and Silu Mu and Xiaolei Qian}, doi = {10.1016/j.asoc.2025.113767}, journal = {Applied Soft Computing}, month = {12}, pages = {113767}, shortjournal = {Appl. Soft. Comput.}, title = {Fine-grained mutation operators for community hiding using genetic algorithms}, volume = {184}, year = {2025}, }
• (2025). Predicting scour depth in the presence of aprons using XGBoost-optuna. ASOC, 184, 113766. (www)

  @article{ , abstract = {This paper explores the impact of an apron in mitigating scour downstream of a trapezoidal PK weir through an experimental study. Three apron lengths were tested with two sediment types under varying hydraulic conditions to address local scouring. Results show that longer aprons reduce scouring, especially at lower densimetric Froude numbers, affecting the location of the maximum scour depth and its volume. On average, apron lengths of 1 P , 1.5 P , and 2 P ( P is weir height) decrease the scour hole areas and volumes by approximately 69–77 %. Scour indices decrease by 73–90 % for corresponding apron lengths. New empirical equations have been proposed to aid in apron design, and the estimation of various scour hole geometries . Bayesian Optimized Neural Network (BONN), Extreme Gradient Boosting model tuned by Optuna algorithm (XGBoost-Optuna), and Random Forest were also developed for forecasting scour hole characteristics in the presence of apron. Various regression tests, including residual plots and uncertainty quantification, were imposed to compare the models. The results demonstrated that the XGBoost-Optuna model outperformed the other models, achieving a correlation coefficient ranging from 0.924 to 0.985, a root mean squared error between 0.055 and 5.072, and a mean relative percentage error of 7.14–11.71 %. Most forecasts generated by the XGBoost-Optuna model fell within ±20 % error margins, highlighting its superiority in predicting scour hole characteristics in the presence of the apron for PK weirs.}, archive = {J_ASOC}, author = {Chonoor Abdi Chooplou and Saeed Balahang and Masoud Ghodsian and Mohammad Vaghefi}, doi = {10.1016/j.asoc.2025.113766}, journal = {Applied Soft Computing}, month = {12}, pages = {113766}, shortjournal = {Appl. Soft. Comput.}, title = {Predicting scour depth in the presence of aprons using XGBoost-optuna}, volume = {184}, year = {2025}, }
• (2025). LanPPT: Enhancing landslide crack detection through pyramid pooling transformers. ASOC, 184, 113765. (www)

  @article{ , abstract = {Landslide cracks, also known as tension cracks, are a major part of landslides. Based on the distribution of landslide fractures, the location of a landslide can be broadly described, and the stress distribution of the sliding mass can be inferred. Cracks at a landslide’s head, which are a key indicator of the displacement of the landslide body, can provide early warning signs for landslide hazards. The complete knowledge of the crack development features is still lacking because of many influencing elements and intricate reasons for fracture creation. Even though some early interventions reported models for automated crack identification utilizing advanced machine learning techniques, the problem still has not been solved to its full potential. Thus, an effective deep architecture for landslide crack segmentation is suggested to address these issues, utilizing a synergistic blend of vision transformers and the pyramid pooling concept. In this work, we use the universal vision transformer backbone called the Pyramid Pooling Transformer, and plug it into our pooling-based multi-head spatial attention to build a deep architecture that identifies and segments the landslide cracks, namely LanPPT. Experiments revealed that when a pyramid pooling transformer is used as the backbone network, it performs significantly better than many earlier convolutional neural networks and normal transformer-based networks in various vision tasks. Systematic experiments show that the proposed model achieved superior performance in terms of mIoU and FPS when compared with the chosen state-of-the-art baselines in the landslide crack detection task.}, archive = {J_ASOC}, author = {S. Sreelakshmi and S.S. Vinod Chandra}, doi = {10.1016/j.asoc.2025.113765}, journal = {Applied Soft Computing}, month = {12}, pages = {113765}, shortjournal = {Appl. Soft. Comput.}, title = {LanPPT: Enhancing landslide crack detection through pyramid pooling transformers}, volume = {184}, year = {2025}, }
• (2025). Differential evolution algorithm with cosine similarity-based individual reduction and symmetric uncertainty-based attribute recovery for feature selection. ASOC, 184, 113764. (www)

  @article{ , abstract = {In recent years, with the increase in data scaling, effectively handling high-dimensional datasets has become a focal point of attention. Feature selection (FS), a method for dealing with datasets containing features, has emerged as a crucial technique in fields such as machine learning and data mining with the objective of selecting features that contain richer information while eliminating redundancy. Due to their remarkable performance in global search, evolutionary computation techniques hold substantial potential in the application of FS. However, many existing FS methods overlook the relationships between features. In the context of classification problems, this study presents a novel wrapper FS algorithm based on the differential evolution algorithm. The proposed method reduces redundant features among individuals based on cosine similarity and selectively and recovers certain features in the crossover phase according to the uncertainty similarity. In addition, a probabilistic-based initialization method is designed. The proposed algorithm significantly outperforms five other algorithms in terms of classification error rates over 18 experimental datasets. The experimental results demonstrate a significant enhancement in the performance of the proposed algorithm attributed to these two components.}, archive = {J_ASOC}, author = {Chunzhi Hou and Ziqian Wang and Yu Zhang and Yuki Todo and Jun Tang and Zhenyu Lei and Shangce Gao}, doi = {10.1016/j.asoc.2025.113764}, journal = {Applied Soft Computing}, month = {12}, pages = {113764}, shortjournal = {Appl. Soft. Comput.}, title = {Differential evolution algorithm with cosine similarity-based individual reduction and symmetric uncertainty-based attribute recovery for feature selection}, volume = {184}, year = {2025}, }
• (2025). Diffusion model based on stochastic differential equation with transformer for stock price prediction. ASOC, 184, 113763. (www)

  @article{ , abstract = {The prediction of stock prices is an exemplary interdisciplinary problem, straddling the domains of finance, computer science, econometrics, and mathematics. The fundamental characteristics of stock price data, notably its non-linearity, non-stationarity, and considerable complexity, make the prediction of stock prices an exceptionally challenging task. In recent years, the domain of deep neural networks has shown substantial promise in terms of learning capacities, leading to significant advancements in the field of stock price prediction. However, most existing methods are still limited to predicting the closing price of the next day, rather than the future trend of stock prices, leaving investors with insufficient information for trading decisions. We propose an innovative model, DiffVT, which incorporates a Volatility Transformer for feature extraction from historical data. We introduced a de-stationary attention mechanism that integrates non-stationary information into the model to capture the dependencies in highly volatile stock price sequences. The output is then fed into an improved diffusion model based on stochastic differential equation (SDE). By iteratively solving the reverse-time SDE, our model generates a probabilistic distribution. This approach not only predicts the single-point stock price for the next day but also forecasts the future trend of stock prices over a period, providing a possible range of stock price movements. To our knowledge, DiffVT is the first model to combine a diffusion model with Transformer for stock price prediction. Extensive experiments on multiple stock indices and individual stock datasets demonstrate that DiffVT significantly outperforms state-of-the-art baseline methods, exhibiting excellent performance across various prediction window lengths. Specifically, compared to the second-best models in each domain, our approach achieves reductions of up to 11.73% in Mean Absolute Error (MAE) for point predictions, 17.23% in Continuous Ranked Probability Score (CRPS) for probabilistic predictions, and an improvement of 12.93% in the Sharpe ratio, clearly establishing its superior performance. Our code is available at https://github.com/SoraKsgn/DiffVT/ .}, archive = {J_ASOC}, author = {Chaoyang Wang and Guangyu Liu and Ling Zhu}, doi = {10.1016/j.asoc.2025.113763}, journal = {Applied Soft Computing}, month = {12}, pages = {113763}, shortjournal = {Appl. Soft. Comput.}, title = {Diffusion model based on stochastic differential equation with transformer for stock price prediction}, volume = {184}, year = {2025}, }
• (2025). Multiobjective coevolutionary bayesian learning for hyperspectral sparse unmixing. ASOC, 184, 113762. (www)

  @article{ , abstract = {The capability of multiobjective evolutionary algorithms (MOEAs) to directly address ℓ 0 norm minimization has introduced innovative perspectives for solving sparse problems. However, enhancing the search efficiency of MOEAs remains a formidable challenge, especially in high-dimensional sparse problems. To alleviate the above problem, we propose a novel multiobjective coevolutionary Bayesian learning framework for a classical sparse problem—sparse unmixing. Leveraging the spatial similarity inherent in hyperspectral image patches, the proposed framework alleviates the complexity of the sparse unmixing task by segmenting the original image into homogeneous regions using superpixel segmentation. These regions are then demixed independently and jointly optimized under a cooperative evolutionary paradigm. During the optimization, the row-sparsity parameter inferred through cooperative Bayesian learning is embedded into a specially designed image-level genetic strategy. This parameter can guide the evolutionary direction of the population, encourages the solution to conform to the structural characteristic and significantly improves search efficiency. Experimental results on synthetic and real datasets demonstrated the effectiveness of the proposed algorithm as compared with several sparse unmixing methods.}, archive = {J_ASOC}, author = {Yiting Liu and Maoguo Gong and Xiangming Jiang and Jianzhao Li and Yue Zhao and Yan Pu and Ziqi Di}, doi = {10.1016/j.asoc.2025.113762}, journal = {Applied Soft Computing}, month = {12}, pages = {113762}, shortjournal = {Appl. Soft. Comput.}, title = {Multiobjective coevolutionary bayesian learning for hyperspectral sparse unmixing}, volume = {184}, year = {2025}, }
• (2025). Real-time anomaly detection in seasonal time series with conditional variational autoencoder. ASOC, 184, 113761. (www)

  @article{ , abstract = {Real-time anomaly detection in high-frequency seasonal time series is commonly addressed using prediction-based methods, which require waiting for new values to perform subsequent predictions and demand continuous processing over time. This work introduces a novel framework for real-time anomaly detection in seasonal time series, with a practical implementation using Conditional Variational Autoencoders based on Multilayer Perceptrons. Our approach eliminates the need for historical time series data at inference time, instead generating a one-shot long-term expected time series that enables immediate evaluation of streaming data with minimal computational resources. Empirical evaluations on real-world seasonal time series demonstrate that the proposed approach achieves state-of-the-art performance compared in both semi-supervised and unsupervised settings. The framework provides computational efficiency and low energy consumption, making it suitable for deployment in commodity hardware and offline environments.}, archive = {J_ASOC}, author = {Lorenzo Porcelli and Marcello Trovati and Francesco Palmieri}, doi = {10.1016/j.asoc.2025.113761}, journal = {Applied Soft Computing}, month = {12}, pages = {113761}, shortjournal = {Appl. Soft. Comput.}, title = {Real-time anomaly detection in seasonal time series with conditional variational autoencoder}, volume = {184}, year = {2025}, }
• (2025). Weight optimized stacked LSTM with conditional random fields using self-adaptive generalized normal distribution optimizer for crop yield forecasting. ASOC, 184, 113760. (www)

  @article{ , abstract = {In the agricultural sector, estimating crop production is a difficult task. A crucial element in recent years has forecasted the crop prediction, which is dependent on outside variables include soil, water, and agricultural characteristics. Crop feature extraction is used to predict crop yield using deep learning-based approaches. The predictive ability of the method heavily relies on the nature of the collected information because there is a nonlinear translation between the unprocessed information and crop yield data. Agricultural marketing, crop production, efficient harvest management, and effective fertilization management rely heavily on crop yield forecasts. Numerous manual analyses are used for remote sensing, which is often used for crop prediction. Here, the deep learning technique is more crucial for predicting crop yields from the remote sensing images, and more complicated approaches are needed to derive the essential spatiotemporal features of the data. Thus, this paper suggests a new weight-optimized Stacked Long Short Term Memory with Conditional Random Field (SLSTM-CRF) and a Modified Self-Adaptive Generalized Normal Distribution Optimizer (MSGNDO) for the prediction of crop yield. This research work has the following phases; data collection, data preprocessing, weighted feature selection, and crop yield forecasting. Primarily, the standard agricultural data is taken from the benchmark sources. Then, data preprocessing is carried out to improve the quality of data. Next, the features are optimally chosen using a newly recommended MSGNDO, in which the weights are tuned via MSGNDO. Further, the tuned weights are multiplied by the extracted features. The final crop yield prediction is done via weight-optimized SLSTM-CRF, where the parameter optimization is done using the same MSGNDO method. The prediction performance of the suggested framework is compared with existing techniques using diverse error-based measures to show effective performance regarding Dataset 1 and Dataset 2. While considering the Root Mean Squared Error (RMSE) analysis, the suggested framework shows 22.4571 less than 29.88774 of CNN, 29.20486 of RNN, 27.9877 of ResNet-50, and 25.38448 of LSTM. Overall, the performance analysis of the developed model while using dataset 1 achieves 3.549 and 24.46 for Mean Absolute Error (MAE) and RMSE, respectively. While comparing dataset 2, the developed model shows values of 5.196 and 29.16 regarding MAE and RMSE measures. Hence, the entire experimental evaluation of the recommended methods shows more effective outcomes than the existing conventional models.}, archive = {J_ASOC}, author = {Shaik Shameer Basha and B S Nissar Begum}, doi = {10.1016/j.asoc.2025.113760}, journal = {Applied Soft Computing}, month = {12}, pages = {113760}, shortjournal = {Appl. Soft. Comput.}, title = {Weight optimized stacked LSTM with conditional random fields using self-adaptive generalized normal distribution optimizer for crop yield forecasting}, volume = {184}, year = {2025}, }
• (2025). An optimized logic mining method for data processing through higher-order satisfiability representation in discrete hopfield neural network. ASOC, 184, 113759. (www)

  @article{ , abstract = {A high performance classification tool such as logic mining has emerged as one of the future computing systems for big data processing. The collaboration between logic and neural network resulted in extracting the most suitable induced logic to represent knowledge from real-life datasets. However, there are certain limitations within the current logic mining models including a non-flexible logical structure, non-optimal computation of the best logic, and the generation of overfitting solutions. Motivated by these limitations, a novel logic mining model incorporating the non-systematic Satisfiability, namely Random 3 Satisfiability in Discrete Hopfield Neural Network is proposed as a logical structure to represent the behaviour of the dataset. The proposed logic mining models used flexible logical structures to prevent overfitting solutions and optimize synaptic weight values. A new computational approach of the best logic by considering True Positive and True Negative values of the learning system is applied in this work to preserve the significance behaviour of the dataset. Furthermore, the comparative experiments of the logic mining models by utilizing various repository real-life datasets are conducted from repositories to assess their efficiency. In accordance with the results, the proposed logic mining model dominates in all the metrics for the average rank. The average rank for each metrics are Accuracy (1.9375), Precision (1.9375), Specificity (1.8125), Mathews Correlation (1.5625), and Fowlkes Mallows Index (2.3125). Numerical results and in-depth analysis demonstrate that the proposed logic mining model consistently produces optimal induced logic that best represents the real-life dataset for all the performance metrics used in this study.}, archive = {J_ASOC}, author = {Nurul Atiqah Romli and Nur Fariha Syaqina Zulkepli and Mohd Shareduwan Mohd Kasihmuddin and Syed Anayet Karim and Siti Zulaikha Mohd Jamaludin and Nur ‘Afifah Rusdi and Gaeithry Manoharam and Mohd. Asyraf Mansor and Nur Ezlin Zamri}, doi = {10.1016/j.asoc.2025.113759}, journal = {Applied Soft Computing}, month = {12}, pages = {113759}, shortjournal = {Appl. Soft. Comput.}, title = {An optimized logic mining method for data processing through higher-order satisfiability representation in discrete hopfield neural network}, volume = {184}, year = {2025}, }
• (2025). Enhancing nonlinear dependencies of mamba via negative feedback for time series forecasting. ASOC, 184, 113758. (www)

  @article{ , abstract = {Mamba is a rising model designed to distill complex patterns from historical data, providing predictive capabilities for time series forecasting tasks. Mamba’s similarity to linear-based models has been criticized due to its limited ability to capture nonlinear dependencies. In this work, we propose a novel model named Embedding C hannel Attention M aclaurin E instein Mamba (CME-Mamba 1 .) based on Mamba framework, with both Embedding Channel Attention and Maclaurin mechanisms incorporated. To further address gradient vanishing issues, we integrate Einstein FFT algorithms, ensuring robust performance against abnormal behaviors of Mamba-based architectures. Extensive experiments conducted on 11 real-world datasets with different numbers of variates, domain focus and granularity, reveal that CME-Mamba achieves state-of-the-art performance in both MSE and MAE, while maintaining reasonable memory efficiency and low time cost. The robustness and credibility of all results are substantiated by a comprehensive convergence and stability analysis. Statistically, consolidated by the Friedman Nonparametric Test and the Wilcoxon Signed-Rank Test, CME-Mamba ranks the first place with significance over counterparts. In addition, in terms of time and memory analysis, CME-Mamba is among the top three models for time and memory efficiency. Despite this, our results further demonstrate that the main contributor is the Embedding Channel Attention Block, which greatly enhances nonlinear dependencies over datasets. The Einstein FFT Block effectively suppresses gradient vanishing occurrences and contributes considerably to performance improvements, driving CME-Mamba both stable and promising. Moreover, the Maclaurin Block based on negative feedback is asymptotically stable without additional gradient vanishing issues and pioneered in achieving synergies with other blocks and greatly enhances nonlinear dependencies. With enhanced nonlinear dependencies generated from the synergy effect of all the three blocks, CME-Mamba grows excellent to uncover complex paradigms and predict future states in various domains, especially improving the performance for periodic and high-variate situations, such as traffic flow management ( ≈ + 8 % ), electricity predictions( ≈ + 6 % ).}, archive = {J_ASOC}, author = {Sijie Xiong and Cheng Tang and Yuanyuan Zhang and Haoling Xiong and Youhao Xu and Atsushi Shimada}, doi = {10.1016/j.asoc.2025.113758}, journal = {Applied Soft Computing}, month = {12}, pages = {113758}, shortjournal = {Appl. Soft. Comput.}, title = {Enhancing nonlinear dependencies of mamba via negative feedback for time series forecasting}, volume = {184}, year = {2025}, }
• (2025). Sequential unsupervised–supervised learning for clustering time-dependent patterns using ellipsoidal calculus. ASOC, 184, 113757. (www)

  @article{ , abstract = {This work introduces a sequential clustering algorithm that combines unsupervised and supervised learning methodologies using ellipsoidal calculus and recurrent neural networks (RNNs). The unsupervised clustering algorithm (ULCA) identifies ellipsoidal sets for the data by applying Lagrange multipliers. These sets are then optimized with gradient descent to adjust their volume and orientation. For time-dependent data, the optimized ellipsoidal sets are updated dynamically by an RNN, which refines their center, orientation, and axis sizes in response to changes in the data. The ULCA is compared to density-based spatial clustering (DBSCAN) and K-means algorithms, showing superior accuracy without the need for pre-determined cluster numbers. Additionally, the convergence of the ULCA and RNN algorithms, working sequentially, is formally proven using Lyapunov stability theory, ensuring continuous classification of data that evolves over time. This study demonstrates the advantages of the proposed hybrid method over traditional clustering algorithms.}, archive = {J_ASOC}, author = {Alejandro Guarneros and Mariana Ballesteros and Iván Salgado and Isaac Chairez}, doi = {10.1016/j.asoc.2025.113757}, journal = {Applied Soft Computing}, month = {12}, pages = {113757}, shortjournal = {Appl. Soft. Comput.}, title = {Sequential unsupervised–supervised learning for clustering time-dependent patterns using ellipsoidal calculus}, volume = {184}, year = {2025}, }
• (2025). Hybrid quantum annealing for large-scale exam scheduling: Validation in real-world educational scenarios. ASOC, 184, 113756. (www)

  @article{ , abstract = {This study applied the hybrid quantum annealing to address complex exam scheduling challenges in large-scale educational scenarios through hybrid quantum annealing enhanced by graph-based preprocessing. By formulating the problem as a Quadratic Unconstrained Binary Optimization (QUBO) model and leveraging the D-Wave Advantage system, our method integrates quantum annealing with classical preprocessing to resolve constraints on exam room availability, student-course conflicts, and batch synchronization. The approach was applied in a university’s 2022 pandemic-era makeup exam scheduling for 1807 students and 215 courses (2749 exam instances) with zero conflicts. Experimental results show that hybrid quantum annealing consumes merely 86 ms of Quantum Processing Unit (QPU) execution time. In contrast, classical simulated annealing requires 13547 ms of Central Processing Unit (CPU) execution time for the same problem scale. This work bridges quantum computing and educational operations, offering a comparative analysis of hybrid algorithms in multi-constraint optimization domains.}, archive = {J_ASOC}, author = {Ziyu Zhou and Qing Chen and Chaojie Zhang and Mengtong Tan and Shuyan Li}, doi = {10.1016/j.asoc.2025.113756}, journal = {Applied Soft Computing}, month = {12}, pages = {113756}, shortjournal = {Appl. Soft. Comput.}, title = {Hybrid quantum annealing for large-scale exam scheduling: Validation in real-world educational scenarios}, volume = {184}, year = {2025}, }
• (2025). Multi-phase evolutionary search space shrinking for large-scale multi-objective feature selection. ASOC, 184, 113755. (www)

  @article{ , abstract = {Feature selection is essential in machine learning, especially for high-dimensional datasets, where irrelevant and redundant features can degrade performance and increase computational cost. In such settings, the search space becomes exponentially large and sparsely populated with relevant solutions, making effective exploration highly challenging. Despite progress in evolutionary methods, many existing algorithms struggle to scale or maintain sparsity. There is an urgent need for scalable strategies that intelligently reduce the search space while retaining essential features. Using feature importance to guide search space shrinking offers a powerful, domain-specific approach tailored for feature selection. This paper proposes a novel large-scale multi-objective evolutionary algorithm, LMSSS, that addresses sparse feature selection through a multi-phase search space shrinking strategy. Features are ranked based on correlation with class labels and frequency in an initial low-cost evolutionary process. A voting-based crossover operator prioritizes parent solutions with better classification accuracy, while a guided mutation mechanism reintroduces prematurely excluded features for reevaluation. These components enable efficient exploration of sparse, high-dimensional spaces. Experiments on 15 large-scale datasets demonstrate that LMSSS achieves superior classification accuracy with smaller, more informative feature subsets compared to state-of-the-art methods, highlighting its effectiveness and scalability.}, archive = {J_ASOC}, author = {Azam Asilian Bidgoli and Shahryar Rahnamayan}, doi = {10.1016/j.asoc.2025.113755}, journal = {Applied Soft Computing}, month = {12}, pages = {113755}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-phase evolutionary search space shrinking for large-scale multi-objective feature selection}, volume = {184}, year = {2025}, }
• (2025). Adaptive information mesh for multimodal brain tumor segmentation. ASOC, 184, 113754. (www)

  @article{ , abstract = {Utilizing advanced deep learning techniques for automatically segmenting brain tumors in multiparametric magnetic resonance imaging (mpMRI) is crucial in enhancing diagnostic processes. However, extreme data imbalance between brain tumors and non-tumorous tissues, as well as among different subregions within the tumors, poses significant challenges for precise tumor segmentation. In this study, we revisit the problem of positive and negative sample balance from the perspective of segmentation difficulty, integrating information entropy theory. We propose an Adaptive Information Mesh that resamples based on the conditional entropy between each pixel in the original data and the final segmentation target. This resampling approach standardizes the data to a uniform level of segmentation difficulty while preserving contextual information. Our methodology includes Multi-density Sparse Convolution and Topological Reconstruction Operator to process data effectively after AI mesh sampling. Multi-density Sparse Convolution exploits the advantages of mesh data to overcome the limitations of traditional sparse convolutions, which cannot exchange information over long distances due to a fixed receptive field. The Topological Reconstruction Operator rebuilds the topological relationships between mesh layers, facilitating information exchange. We conducted training on the BraTS 2018, 2019, and 2021 datasets. Comprehensive benchmark testing on the same metrics demonstrates that our performance is comparable to state-of-the-art structured networks in terms of accuracy while reasonably allocating computational resources. Results on the ISLES 2022 dataset for infarct segmentation in ischemic stroke underscore the robustness of our method.}, archive = {J_ASOC}, author = {Qingfan Hou and Yanjun Peng and Zhuofei Wang and Jian Jiang and Nan Lv}, doi = {10.1016/j.asoc.2025.113754}, journal = {Applied Soft Computing}, month = {12}, pages = {113754}, shortjournal = {Appl. Soft. Comput.}, title = {Adaptive information mesh for multimodal brain tumor segmentation}, volume = {184}, year = {2025}, }
• (2025). An attention-based multiple band interaction network for motor imagery EEG decoding. ASOC, 184, 113750. (www)

  @article{ , abstract = {Brain-computer interfaces (BCIs) based on motor imagery electroencephalogram (MI-EEG) have been extensively used in many applications to assist disabled people. Key frequency bands play a crucial role in MI-EEG signal decoding. However, existing researches pay insufficient attention to the interaction among different bands, especially during feature extraction process. To address this issue, in this work, a multiple band interaction network, called MBINet, is carefully constructed, which adopts band-dependent multi-branch setting. Particularly, an attention-based guide block and a multi-scale interaction block are elaborately designed. The former allows to improve the feature extraction process of μ and β bands by the lights of full band characteristics, and the latter enables to promote multi-angle and multi-scale interactive fusion of high-order significant features from multiple bands. The performance testing of MBINet is performed on two publicly available MI datasets, namely the BCI competition IV-2a dataset and the High gamma dataset. The experimental results show that MBINet outperforms the state-of-the-art methods, with an average classification accuracy of 81.66% and 95.78%, respectively. MBINet provides a new perspective for decoding nonlinear time series, especially EEG signals, by emphasizing diverse interactions between key frequency bands.}, archive = {J_ASOC}, author = {Weidong Dang and Kefa Zhang and Haoyu Li and Dongmei Lv and Wei Guo and Zhongke Gao}, doi = {10.1016/j.asoc.2025.113750}, journal = {Applied Soft Computing}, month = {12}, pages = {113750}, shortjournal = {Appl. Soft. Comput.}, title = {An attention-based multiple band interaction network for motor imagery EEG decoding}, volume = {184}, year = {2025}, }
• (2025). The neighborhood rough set based on division-mining-fusion strategy. ASOC, 184, 113749. (www)

  @article{ , abstract = {The neighborhood rough set (NRS) model exhibits significant value in numerous data mining tasks. However, most existing NRS models are only suitable for data mining problems in a single dataset, and their comprehensive performance in processing data is not ideal. To address these two challenges, we develop a novel NRS model by applying the division-mining-fusion (DMF) strategy. Specifically, we first decompose the original dataset into multiple subsets. Then, a submodel is established for each subset, and finally, all submodels are integrated to develop a new NRS model. The experimental results demonstrate that the developed model can not only effectively handle the data mining tasks in scenarios with multiple datasets, but also has outstanding comprehensive performance. The NRS model developed in the paper provides a novel solution for efficiently processing large-scale complex data.}, archive = {J_ASOC}, author = {Conghao Yan and Qingzhao Kong and Wenbin Zhang}, doi = {10.1016/j.asoc.2025.113749}, journal = {Applied Soft Computing}, month = {12}, pages = {113749}, shortjournal = {Appl. Soft. Comput.}, title = {The neighborhood rough set based on division-mining-fusion strategy}, volume = {184}, year = {2025}, }
• (2025). Explainable deep learning framework for lower limb activity classification using reconstructed sEMG signals in healthy and pathological subjects. ASOC, 184, 113748. (www)

  @article{ , abstract = {Recognizing human activities from Surface Electromyography (sEMG) signals is fraught with challenges, including noise susceptibility and signal crosstalk. Addressing these, our study analyses sEMG data from 11 healthy and 11 pathological subjects across three distinct activities: sitting, standing, and walking. We introduce a pioneering preprocessing approach that integrates Bandpass Filtering, Wavelet Denoising, and Ensemble Empirical Mode Decomposition (EEMD) for signal enhancement. A significant aspect of our approach involves the reconstruction of signals by selecting the top 50% of Intrinsic Mode Functions (IMFs) based on entropy, Signal-to-Noise Ratio (SNR), correlation, and energy, effectively capturing the most informative features of the sEMG signals. To counterbalance dataset imbalances and facilitate robust feature extraction, we applied Adaptive Synthetic (ADASYN) sampling and segmented the data into 256 ms windows with a 25% overlap. Our Convolution Neural Network (CNN) achieves remarkable classification accuracies: for sitting, 99.4% in healthy and 99.2% in pathological subjects; for standing, 99.7% and 98.8% respectively; and for walking, 99.4% and 98.6%, respectively. Furthermore, the integration of Explainable AI (XAI) through Permutation Feature Importance (PFI) provides critical insights into the significant impact of muscle signals, particularly highlighting the Rectus Femoris (RF) muscle’s role in sitting with leg extension, BF muscles’s role in standing with flexion. This comprehensive and innovative methodology not only overcomes the inherent challenges of sEMG signal analysis but also enhances the interpretability and reliability of activity recognition, marking a significant advancement for personalized healthcare interventions.}, archive = {J_ASOC}, author = {Pratibha Tokas and Vijay Bhaskar Semwal and Sweta Jain}, doi = {10.1016/j.asoc.2025.113748}, journal = {Applied Soft Computing}, month = {12}, pages = {113748}, shortjournal = {Appl. Soft. Comput.}, title = {Explainable deep learning framework for lower limb activity classification using reconstructed sEMG signals in healthy and pathological subjects}, volume = {184}, year = {2025}, }
• (2025). Mission assignment and path planning for ground forces using permutation-based simplified swarm optimization with Q-learning adaptation. ASOC, 184, 113747. (www)

  @article{ , abstract = {This study tackles task assignment and path planning for counteroffensive operations using a bi-level programming model inspired by spatial crowdsourcing. While existing models often prioritize centralized platform decisions, they tend to overlook the autonomy and real-time judgment of on-site participants, whose local actions critically affect coordination outcomes. To address this, we propose the Defender Deployment and Routing Model (DDRM), which integrates upper-level task assignment with lower-level route planning. To solve DDRM efficiently, we develop a hybrid nested framework integrating an Adaptive Permutation-Based Simplified Swarm Optimization (APSSO) algorithm with Dijkstra’s algorithm. APSSO incorporates a task-aware initialization strategy and a lightweight Q-learning mechanism for operator selection. As the first SSO-based approach with permutation-based operators, APSSO effectively handles discrete task allocation structures. Experimental results on 36 benchmark instances show that APSSO achieves the best performance in 34 cases compared to six tailored evolutionary variants. Statistical analysis confirms its superiority in solution quality, while the added Q-learning module introduces minimal computational overhead. These results highlight APSSO’s efficiency and robustness across varying problem scales.}, archive = {J_ASOC}, author = {Jun-Lin Lin and Chyh-Ming Lai and Song-Pei Wu}, doi = {10.1016/j.asoc.2025.113747}, journal = {Applied Soft Computing}, month = {12}, pages = {113747}, shortjournal = {Appl. Soft. Comput.}, title = {Mission assignment and path planning for ground forces using permutation-based simplified swarm optimization with Q-learning adaptation}, volume = {184}, year = {2025}, }
• (2025). Dynamic event-triggered adaptive fuzzy admittance control of robotic systems with uncertainties. ASOC, 184, 113746. (www)

  @article{ , abstract = {In modern industrial production and daily life, the increasing complexity of application scenarios has led to higher requirements for the compliance and safety of robotic systems. However, during robot control, a large number of redundant signals are often sampled, which significantly increases the communication burden. Therefore, how to substantially reduce the communication burden while ensuring satisfactory performance has become an urgent issue to be addressed. To address this challenge, this paper proposes a dynamic event-triggered adaptive fuzzy admittance control (DETAFAC) strategy for robotic systems with uncertainties, where the more aggressive dynamic event-triggered condition can significantly reduce the communication burden and the admittance model is used to reshape the desired trajectory of the robotic systems. Additionally, a fuzzy logic system (FLS) is utilized to address the uncertainties of the robotic systems, the update law of the FLS and the stability of the control system are examined using the Lyapunov stability theorem, and the dynamic triggering condition is formulated to prevent Zeno behavior. Simulation and experimental validations are performed, and the results demonstrate that the proposed DETAFAC strategy can achieve better performances in comparison to the similar approaches.}, archive = {J_ASOC}, author = {Jinzhu Peng and Xuxin Liu and Shuai Ding and Yaqiang Liu}, doi = {10.1016/j.asoc.2025.113746}, journal = {Applied Soft Computing}, month = {12}, pages = {113746}, shortjournal = {Appl. Soft. Comput.}, title = {Dynamic event-triggered adaptive fuzzy admittance control of robotic systems with uncertainties}, volume = {184}, year = {2025}, }
• (2025). Boxing action recognition using inertial data and deep learning. ASOC, 184, 113745. (www)

  @article{ , abstract = {This study explores the ongoing shift toward automated sports performance analysis by employing Deep Convolutional Neural Networks (DCNNs) for the classification of movement data in combat sports, specifically boxing. Using data from wearable inertial measurement units (IMUs), this study addresses the limitations of traditional, expensive video analysis by providing an accessible, cost-effective alternative that can be seamlessly integrated into athletes' training routines. IMUs offer a practical solution for continuous performance monitoring and feedback, establishing a foundation for automated performance assessment during training sessions. In this study, the classification accuracy and recall of DCNNs are rigorously compared with alternative algorithms, demonstrating a high overall recall of 99 % and accuracy of 91 %. Focusing on pad work training as a key application area, this work advances the automation of boxing action recognition, with implications for both training optimization and real-time scoring automation in competitive boxing. These findings underscore the potential for wearable technology and advanced machine learning methods to transform athletic performance evaluation and bring engineering innovation to sports training.}, archive = {J_ASOC}, author = {J. Brindha and G. Nallavan and Radana Vilimkova Kahankova and Radek Martinek}, doi = {10.1016/j.asoc.2025.113745}, journal = {Applied Soft Computing}, month = {12}, pages = {113745}, shortjournal = {Appl. Soft. Comput.}, title = {Boxing action recognition using inertial data and deep learning}, volume = {184}, year = {2025}, }
• (2025). Adaptive dual-network control for modular robots in human–robot interaction. ASOC, 184, 113744. (www)

  @article{ , abstract = {In order to address safety and compliance challenges in physical human–robot interaction (pHRI), this paper presents an adaptive dual-neural network (NN) control framework for modular robot manipulators (MRMs). The proposed approach enables seamless switching between autonomous trajectory tracking and compliant motion guided by human interaction. An adaptive fuzzy model uncertainty compensation method is employed to estimate the interaction torque that is utilized to detect physical contact, determining the MRM’s operation mode. Additionally, the reference trajectory tracking information is provided by human motion intention identification via NNs. This method avoids the need for external force/torque sensors and increases the intention recognition accuracy by jointly estimating human impedance parameters. The overall control scheme simultaneously incorporates adaptive fuzzy PD control and decentralized impedance control through a mode-switching structure, ensuring compliant and accurate motion in the two operation modes. A Lyapunov-based analysis is then conducted. The obtained results demonstrate that the tracking errors of the closed-loop system are uniformly ultimately bounded (UUB). An experimental validation for various tasks also demonstrates the high robustness and effectiveness of the proposed strategy. Note that all the procedures adopted in this paper are conducted following standard safety guidelines, with full consideration of ethical aspects in pHRI.}, archive = {J_ASOC}, author = {Yuexi Wang and Tianjiao An and Bo Dong and Mingchao Zhu and Yuanchun Li}, doi = {10.1016/j.asoc.2025.113744}, journal = {Applied Soft Computing}, month = {12}, pages = {113744}, shortjournal = {Appl. Soft. Comput.}, title = {Adaptive dual-network control for modular robots in human–robot interaction}, volume = {184}, year = {2025}, }
• (2025). Dynamic-informed machine learning: Data-driven reconstruction of delay differential equations models. ASOC, 184, 113743. (www)

  @article{ , abstract = {To enhance the understanding of the dynamical mechanisms underlying complex systems, this study presents a novel data-driven framework for modeling Delay Differential Equations. The Galerkin-Koornwinder theory is combined with Neural Ordinary Differential Equations for the first time, addressing the significant challenge of reconstructing models that incorporate time-delay effects by approximating them with Ordinary Differential Equations within a rigorous mathematical framework. Leveraging Neural Ordinary Differential Equations to learn these approximate models, the approach achieves high interpretability and generalizability, effectively capturing the dynamics of various delay systems, including those with discrete delays, multi-time delays, and distributed delays. Extensive simulation experiments on models exhibiting bifurcation and chaos, as well as real-world biological data, demonstrate the superior performance of this method in long-term prediction and system dynamics reconstruction. When validated on a set of real-world biological experimental data, the long-term behavior prediction error was reduced from 11% to below 1%. The comparison with similar studies elucidates that the approach can accurately capture complex behaviors and outperforms existing methods in terms of prediction accuracy and generalization capability. All data and code are openly available, facilitating reproducibility and further research.}, archive = {J_ASOC}, author = {Xiyuan Chen and Zhong Liu and Qiubao Wang and Zikun Han}, doi = {10.1016/j.asoc.2025.113743}, journal = {Applied Soft Computing}, month = {12}, pages = {113743}, shortjournal = {Appl. Soft. Comput.}, title = {Dynamic-informed machine learning: Data-driven reconstruction of delay differential equations models}, volume = {184}, year = {2025}, }
• (2025). Fusion of data-driven models with a knowledge-guided loss function for flood forecasting. ASOC, 184, 113742. (www)

  @article{ , abstract = {Heavy rainfall frequently causes flooding disasters in basins of all sizes; however, small and medium-sized basins -typically defined as drainage areas less than 1000 km²- are particularly vulnerable due to their rapid runoff response, often resulting in casualties and property losses. Researchers worldwide have made excellent progress in advancing flood prediction capabilities by developing intelligent data-driven models. However, these models rely only on available data and ignore expert knowledge of floods and fail to account for the varying importance of prediction errors during flood and non-flood events in their training process. Therefore, we propose fusing data-driven models with expert knowledge frameworks by designing a custom loss function for accurate flood forecasting. The proposed model effectively interprets the complexity of flood events in small and medium-sized basins and introduces accurate results. By incorporating expert knowledge constraints, it emphasises the importance of prediction errors during critical flood events. The model was tested using hourly streamflow data from the Heihe and Tunxi basins in China. Results show that our model improves prediction accuracy by up to 80 % in root mean square error (RMSE) and 87 % in mean absolute error (MAE) in Heihe and increases the coefficient of determination (R²) by 66 % and the Kling–Gupta efficiency (KGE) by 58 % in Tunxi compared to baseline models such as INFORMER, TD-CNN-LSTM, STA-LSTM, STA-TCN, LSTM, CNN, TCN, and SVR. These findings demonstrate the model’s superior performance in providing accurate and reliable 6-hour-ahead (T + 6) flood forecasts.}, archive = {J_ASOC}, author = {Haider Malik and Jun Feng and Mohammed Abdallah and Jiru Zhang and Pingping Shao and Zaid Ameen Abduljabbar}, doi = {10.1016/j.asoc.2025.113742}, journal = {Applied Soft Computing}, month = {12}, pages = {113742}, shortjournal = {Appl. Soft. Comput.}, title = {Fusion of data-driven models with a knowledge-guided loss function for flood forecasting}, volume = {184}, year = {2025}, }
• (2025). Self-organizing interval type-2 fuzzy neural network based on eigenvalue decomposition. ASOC, 184, 113741. (www)

  @article{ , abstract = {In this study, an eigenvalue decomposition-based self-organizing interval type-2 fuzzy neural network (ED-SOIT2FNN) is proposed to tackle the identification problem of nonlinear systems. The network model determines both the structure and parameters of the network through online learning, which can realize the structure learning and parameter learning simultaneously. Firstly, in the structure learning process of ED-SOIT2FNN, the error criterion and the completeness criterion of fuzzy rules are used to verify whether the rules grow. Meanwhile, the eigenvalue decomposition method is adopted to find the less active rules for deletion, so that obtain a more compact network structure. Secondly, in terms of ED-SOIT2FNN parameter optimization and the characteristics of network parameters, they are divided into the linear and nonlinear ones. The algorithm of adaptive discount recursive partial least square is employed to optimize the linear parameters, which is conducive to improving the noise resistance of the network model and solving the data saturation problem. And the sliding window adaptive second-order algorithm with a forgetting factor is adopted to optimize the nonlinear parameters. Compared with the algorithm of gradient descent optimization, it can accelerate the convergence and achieve a good adaptability with stability. Finally, the proposed ED-SOIT2FNN was applied to four typical nonlinear examples for identification. The experimental results showed that compared with similar methods in the existing literature, the proposed ED-SOIT2FNN could produce a more compact network structure with higher accuracies of identification and prediction.}, archive = {J_ASOC}, author = {Panchao Wang and Taoyan Zhao and Jiangtao Cao and Ping Li}, doi = {10.1016/j.asoc.2025.113741}, journal = {Applied Soft Computing}, month = {12}, pages = {113741}, shortjournal = {Appl. Soft. Comput.}, title = {Self-organizing interval type-2 fuzzy neural network based on eigenvalue decomposition}, volume = {184}, year = {2025}, }
• (2025). Berth-quay crane-experiment allocation method based on improved genetic algorithm for cargo and scientific research port. ASOC, 184, 113740. (www)

  @article{ , abstract = {Against the background of booming global shipping trade and growing demand for marine scientific research, multifunctional ports with both cargo transportation and scientific research functions, namely CSR_Ports, are facing significant operational scheduling challenges. Due to the intricate mutual influence and resource competition among cargo vessels, research vessels, and marine experiments, the existing method struggles to cope with the diversified demands of port operations. To solve the above challenges, this study considers the differences in infrastructure requirements between cargo vessels and research vessels, designs differentiated work areas, abstracts the berth-quay crane-experiment allocation problem into a two-dimensional packing problem with time window constraints, and establishes a dual-objective berth-quay crane-experiment allocation (BQCEA) model to minimize the vessels' average turnaround time and the average variation of the experiment start time. Then, given the limitations of genetic algorithms, such as insufficient adaptive ability and a tendency to fall into local optimality, an improved algorithm, namely NQRGA, is proposed based on neighborhood search, quantum multi-point crossover, and retention mechanism. Finally, the BQCEA model is combined with the NQRGA to propose a novel solution method for solving the berth-quay crane-experiment allocation problem, referred to as BQCEA_NQRGA. The effectiveness of the BQCEA_NQRGA method is evaluated using real operational data from two CSR_Ports in China. The test results show that, compared with the FCFS method, the BQCEA_NQRGA method achieves at least a 49.79 % performance improvement, and the optimization effect gradually increases to 101 % with the increase in the scheduling scale. Compared with the six advanced algorithms selected in this paper, the proposed NQRGA performs best in all six different instances. Notably, the BQCEA_NQRGA method can generate high-quality allocation schemes for 70 vessels and 14 experiments simultaneously within 100 s.}, archive = {J_ASOC}, author = {Ming-Wei Li and Xiang-Yang Li and Jing Geng and Zhong-Yi Yang and Wei-Chiang Hong}, doi = {10.1016/j.asoc.2025.113740}, journal = {Applied Soft Computing}, month = {12}, pages = {113740}, shortjournal = {Appl. Soft. Comput.}, title = {Berth-quay crane-experiment allocation method based on improved genetic algorithm for cargo and scientific research port}, volume = {184}, year = {2025}, }
• (2025). A novel deep reinforcement learning model based on DDPG considering attention mechanism and combined with GRU network for short-term load forecasting. ASOC, 184, 113739. (www)

  @article{ , abstract = {Short-term load forecasting plays a crucial role in power system operations and energy market efficiency. This paper addresses the challenge of capturing complex nonlinear patterns and temporal dependencies in electricity load data, which traditional forecasting methods often fail to handle effectively. We propose a novel deep reinforcement learning approach that combines attention mechanisms with gated recurrent units within a deep deterministic policy gradient framework (Attention-GRU-DDPG). The key innovation lies in treating load forecasting as a decision-making problem, where the model learns to adapt its predictions based on multivariate inputs including historical load, weather conditions, and electricity prices. Our approach uniquely integrates three complementary components: attention mechanisms for feature prioritization, GRU networks for temporal pattern recognition, and reinforcement learning for adaptive strategy optimization. Extensive experiments on Australian electricity market data demonstrate that our model achieves superior forecasting accuracy compared to ten benchmark methods. The proposed framework offers a practical solution for power system operators requiring accurate 24–168 h ahead load predictions, contributing to more efficient grid management and market operations.}, archive = {J_ASOC}, author = {Xin He and Wenlu Zhao and Zhijun Gao and Licheng Zhang and Qiushi Zhang and Xinyu Li}, doi = {10.1016/j.asoc.2025.113739}, journal = {Applied Soft Computing}, month = {12}, pages = {113739}, shortjournal = {Appl. Soft. Comput.}, title = {A novel deep reinforcement learning model based on DDPG considering attention mechanism and combined with GRU network for short-term load forecasting}, volume = {184}, year = {2025}, }
• (2025). A novel blood supply management model using evolutionary neural network. ASOC, 184, 113738. (www)

  @article{ , abstract = {Increasing demand and supply uncertainties and the short lifetime of blood products cause wastage of the blood gathered from donors. Also, there is a high shortage of blood products because of the limited number of donors and emergency demands. Hence, there is a great significance to constructing a blood supply management model by minimizing (1) transportation cost, (2) shortage cost, (3) wastage cost, (4) ordering cost, and (5) inventory holding cost. The different types of costs with constraints make blood supply management a challenging multi-objective combinatorial optimization problem. Earlier, neural genetic algorithms mostly concentrate on single-objective optimization problems. In this paper, we propose a blood supply management procedure using a network-driven evolutionary network (D2NNEA). A set of pointer networks are utilized to solve the multi-objective problem. The performance of the proposed technique is analyzed by comparing it with existing techniques and the superiority of the proposed technique in solving multi-objective optimization problems is verified.}, archive = {J_ASOC}, author = {Harinandan Tunga and Soumyadip Dhar and Samarjit Kar and Debasis Giri}, doi = {10.1016/j.asoc.2025.113738}, journal = {Applied Soft Computing}, month = {12}, pages = {113738}, shortjournal = {Appl. Soft. Comput.}, title = {A novel blood supply management model using evolutionary neural network}, volume = {184}, year = {2025}, }
• (2025). Partial unified learning for dynamic change detection in hyperspectral images. ASOC, 184, 113737. (www)

  @article{ , abstract = {Hyperspectral image change detection (HSI-CD) aims to identify changes in bi-temporal hyperspectral images (HSIs) captured at different times in the same location. Existing algorithms often overlook the inherent class imbalance in HSI-CD, leading to poor generalization in detecting changes while introducing redundant computation in unchanged regions. This paper introduces a novel mechanism based on Partial Unified Learning for Dynamic Change Detection (PUL-DCD) to address these limitations. Particularly, a novel partial unified learning network is proposed, whose backbone is trained using multiple datasets, whilst the task-specific networks are trained independently with each individual dataset. In so doing, the network can maintain outstanding performance on specific datasets while having strong generalization ability. Furthermore, an innovative dynamic architecture is introduced that distinguishes between easy and hard regions for change detection, thereby optimizing parameter configuration and enhancing detection performance in challenging areas, while mitigating redundancy regarding unchanged information. Experimental results on three datasets show that PUL-DCD is competitive in both accuracy and efficiency.}, archive = {J_ASOC}, author = {Keyun Zhao and Ying Li and Qingping Zheng and Qiang Shen}, doi = {10.1016/j.asoc.2025.113737}, journal = {Applied Soft Computing}, month = {12}, pages = {113737}, shortjournal = {Appl. Soft. Comput.}, title = {Partial unified learning for dynamic change detection in hyperspectral images}, volume = {184}, year = {2025}, }
• (2025). Offline reinforcement learning for job-shop scheduling problems. ASOC, 184, 113736. (www)

  @article{ , abstract = {Recent advances in deep learning have shown significant potential for solving combinatorial optimization problems in real-time. Unlike traditional methods, deep learning can generate high-quality solutions efficiently, which is crucial for applications like routing and scheduling. However, existing approaches like deep reinforcement learning (RL) and behavioral cloning have notable limitations, with deep RL suffering from slow learning and behavioral cloning relying solely on expert actions, which can lead to generalization issues and neglect of the optimization objective. Offline RL addresses these challenges by learning from fixed datasets while leveraging reward signals, making it especially suitable for constrained combinatorial problems where online exploration is impractical. This paper introduces a novel offline RL method designed for combinatorial optimization problems with complex constraints, where the state is represented as a heterogeneous graph and the action space is variable. Our approach encodes actions in edge attributes and balances expected rewards with the imitation of expert solutions. We demonstrate the effectiveness of this method on job-shop scheduling and flexible job-shop scheduling benchmarks, achieving superior performance compared to state-of-the-art techniques.}, archive = {J_ASOC}, author = {Imanol Echeverria and Maialen Murua and Roberto Santana}, doi = {10.1016/j.asoc.2025.113736}, journal = {Applied Soft Computing}, month = {12}, pages = {113736}, shortjournal = {Appl. Soft. Comput.}, title = {Offline reinforcement learning for job-shop scheduling problems}, volume = {184}, year = {2025}, }
• (2025). Surrogate-assisted multi-objective covariance matrix adaptation evolution strategies. ASOC, 184, 113728. (www)

  @article{ , abstract = {Although the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is one of the most preferred evolutionary algorithms for single-objective black-box optimization, its performance is limited in the multi-objective space due to its reliance solely on Gaussian-based mutation without any crossover for offspring generation. To address this limitation, this work proposes a surrogate-assisted multi-objective CMA-ES algorithm with an ensemble of offspring generation schemes. In the proposed algorithm, trial solutions are generated from an ensemble of the standard CMA-ES operator and a Genetic Algorithm-inspired operator. Consequently, the solutions are evaluated on a Gaussian Process-based surrogate model, and the solution with the best Expected Improvement (EI) is selected as the generated offspring. Experiments on the Walking Fish Group (WFG) test suite and 18 benchmark multi-objective Neural Architecture Search (NAS) problems demonstrate that the proposed approach is statistically superior to existing multi-objective CMA-ES variants and other state-of-the-art non-CMA-ES multi-objective algorithms. Specifically, the proposed algorithm achieves a win rate of 79.63% and 77.8% on the WFG and NAS test suites, respectively, against other CMA-ES variants, and demonstrates a 68.8% win rate against state-of-the-art algorithms on the NAS test suite.}, archive = {J_ASOC}, author = {Oladayo S. Ajani and Adeyinka Adedigba and Kalyana C. Veluvolu and Rammohan Mallipeddi}, doi = {10.1016/j.asoc.2025.113728}, journal = {Applied Soft Computing}, month = {12}, pages = {113728}, shortjournal = {Appl. Soft. Comput.}, title = {Surrogate-assisted multi-objective covariance matrix adaptation evolution strategies}, volume = {184}, year = {2025}, }
• (2025). A multi-source domain adaptation approach with learning domain-specific representations for bearing fault diagnosis under limited samples. ASOC, 184, 113727. (www)

  @article{ , abstract = {Bearings, vital to train traction motors, directly influence train safety due to their exposure to harsh environments and alternating loads, leading to various faults. Bearing fault diagnosis is thus critical for ensuring train safety. However, the challenge of data privacy and the scarcity of data make diagnosing faults difficult. Currently, transfer models based on single-source domain adaptation are used to address the above issues. However, data from multiple related domains may exist simultaneously in some scenarios. Therefore, learning fault knowledge from multiple domains and transferring it to the target domain for faults is a difficult problem. In view of this, we propose a multi-source domain adaptation approach with learning domain-specific representations (MDA-LDDSR) for bearing fault diagnosis under limited samples. First, MDA-LDDSR adds a domain-specific feature extractor (MAS-Net) for source and target domain to align the target domain with the distribution. After purifying the features by a partial feature selection strategy, we innovatively construct an intra-class alignment strategy to keep the marginal and conditional distributions of the data in alignment. Then, we utilize the instance-to-domain Mahalanobis distance to explicitly model the similarity between different domains. Finally, this similarity is assigned in the form of weights to multiple domain hypotheses. Superiorly, experiments in real rolling bearing fault diagnosis also show that the model has better convergence while it is characterized by high efficiency and robustness. The proposed approach has a strong advantage to be used for real-time bearing fault diagnosis when noise is added to the environment. Also, it obtains the desired diagnosis results compared to other approaches of the same type.}, archive = {J_ASOC}, author = {Qiang Zhou and Wengang Ma and Yadong Zhang and Jin Guo}, doi = {10.1016/j.asoc.2025.113727}, journal = {Applied Soft Computing}, month = {12}, pages = {113727}, shortjournal = {Appl. Soft. Comput.}, title = {A multi-source domain adaptation approach with learning domain-specific representations for bearing fault diagnosis under limited samples}, volume = {184}, year = {2025}, }
• (2025). Surrounding and context-aware network for individual emotion complement in social networks. ASOC, 184, 113726. (www)

  @article{ , abstract = {Most sentiment analysis focuses on identifying individual emotional shifts related to a specific topic or predicting an individual’s emotional state in the future. However, it overlooks the influence of surrounding factors and contextual emotional information within social networks. Furthermore, it is also important to address the gap in understanding the missing emotions of individuals during critical historical moments. In this paper, we propose a surrounding and context-aware network for individual emotion complement(denoted as SCAEC) to infer missing emotional states at pivotal moments in history. The SCAEC network consists of an encoder layer for emotion embedding, a surrounding-aware layer, a context-aware layer, a feature interactive learning layer and an emotion distribution layer. Additionally, we introduce a decay long short-term memory (LSTM) network (DLSTM) within the SCAEC, which effectively extracts emotional features from individuals during moments of emotional absence and allows for the decay of emotional influence over time. We use DLSTM to extract associated emotional features as global features and apply multi-head attention to capture key emotional features as local features from posts/comments before and after moments of emotional absence. This allows us to obtain surrounding and context features based on the global and local features, respectively. The feature interactive learning layer then combines these features between the surrounding and context features to learn their interaction. Additionally, the emotion distribution layer simulates the emotional interaction between the individual and their surroundings. Finally, we derive the individual’s emotion distribution. Experimental results demonstrate that SCAEC model achieves an F1 score of 67.52% on the X and 68.98% on the Microblog dataset, marking an improvement of 1%–10% compared to baseline methods.}, archive = {J_ASOC}, author = {Sai Kang and YaJun Du and Xianyong Li and Xiaoliang Chen and Chunzhi Xie and Jia Liu and Yan-li Lee}, doi = {10.1016/j.asoc.2025.113726}, journal = {Applied Soft Computing}, month = {12}, pages = {113726}, shortjournal = {Appl. Soft. Comput.}, title = {Surrounding and context-aware network for individual emotion complement in social networks}, volume = {184}, year = {2025}, }
• (2025). Spam bot detection on twitter platform using positional attention based dense convolutional neural network. ASOC, 184, 113725. (www)

  @article{ , abstract = {Nowadays, social media platforms such as Twitter and Facebook play a major role in everyday life because of the incredible opportunities they offer users. Twitter is the most essential social media platform since it allows people to express themselves through tweets. However, because of its popularity among a vast number of users, the Twitter platform is being abused by automated accounts known as bots. Since automated accounts transmit fake news, fake ideas, and fake products, early detection of bots on the Twitter platform is critical. Previously, researchers introduced ineffective methods for identifying social media bots. Positional Attention-based Dense Convolutional Neural Network (PAtt_Dense CNN) is a new deep learning-based spam bot detection framework proposed in this paper. Pre-processing of the incoming data includes stop word removal, tokenization, stemming, n-gram identification, user mention, URL and hashtag removal, vocabulary density, and richness analysis. From the pre-processed images, significant features are extracted utilizing N-gram level vectorizer, TF-IDF vectorizer, character level vectorizer, and Extended Bidirectional Encoder Representations from Transformer (EBERT). After feature extraction, binary water wheel plant optimization is used to select the best characteristics. Finally, spam bot detection and classification are performed using a PAtt_Dense CNN. The performance of the proposed technique is then evaluated by analyzing the performance indicators and comparing them to existing procedures. According to the comparison results, the proposed technique achieved the enhanced outcome in terms of Accuracy, Precision, Recall, and F1-Score of 97.8 %, 97.2 %, 98.3 %, and 99.2 %, respectively.}, archive = {J_ASOC}, author = {Hemal Girishkumar Shah and Hiren Joshi}, doi = {10.1016/j.asoc.2025.113725}, journal = {Applied Soft Computing}, month = {12}, pages = {113725}, shortjournal = {Appl. Soft. Comput.}, title = {Spam bot detection on twitter platform using positional attention based dense convolutional neural network}, volume = {184}, year = {2025}, }
• (2025). Consistency- and ordinal consensus-based multiple scenario model for group decision-making with incomplete probabilistic linguistic preference relations. ASOC, 184, 113722. (www)

  @article{ , abstract = {Incomplete probabilistic linguistic preference relations (InPLPRs), as a practical expression to portray the uncertainty of things, can describe the information on decision makers’ (DMs’) evaluation of things in pairwise comparisons in group decision making with two dimensions simultaneously. This paper investigates a group decision-making (GDM) method with InPLPRs to express the preference information of DMs and establish a consensus reaching mechanism for multiple scenarios. First, to obtain incomplete information, we define the concept of opinion swing neighborhood based on referring to the opinions of others and analyzing the mental behavior of DMs. Furthermore, a missing information estimation model of InPLPRs considering the opinion swing neighborhood is developed to obtain consistent optimal estimates. Secondly, the ordinal consensus index is defined, facilitating a more accurate measure of consensus attainment. Then, we thoroughly explore the classification of decision situations using consistency, consensus, illogical rate, and distinguishing index. Based on this, the corresponding consensus optimization models are proposed for different decision scenarios. Subsequently, the DMs’ weights are presented for aggregating individual opinions with acceptable consistency and consensus into group opinions, and the collective opinions are ranked and selected. Finally, numerical examples, simulation experiments, and comparative analysis demonstrate the proposed method’s applicability, effectiveness, and advantages. The proposed method offers a comprehensive GDM approach based on InPLPRs, which can be applied to various real-world GDM scenarios, demonstrating broad applicability.}, archive = {J_ASOC}, author = {Ran Dang and Peide Liu and Peng Wang and Luis Martínez}, doi = {10.1016/j.asoc.2025.113722}, journal = {Applied Soft Computing}, month = {12}, pages = {113722}, shortjournal = {Appl. Soft. Comput.}, title = {Consistency- and ordinal consensus-based multiple scenario model for group decision-making with incomplete probabilistic linguistic preference relations}, volume = {184}, year = {2025}, }
• (2025). Equidistant deep embedding-based multi-label group activity recognition with dependency-constrained training. ASOC, 184, 113721. (www)

  @article{ , abstract = {Group activities in real-world scenarios are often complex and require multiple labels for adequate representation, as they involve multifaceted social interactions among individuals within a small group setting. Existing group activity recognition (GAR) methods primarily focus on model inference, overlooking the inherent challenges of multi-label learning (MLL). One common approach to MLL is the label powerset (LP) method, which transforms multi-hot label vectors into one-hot vectors, converting the multi-label problem into a single-label multi-classification problem. However, this projection loses correlations among active elements in multi-hot vectors, hindering the capture of inter-activity dependencies. To address this limitation, we propose a novel equidistant deep embedding-based (EDE) multi-label GAR framework with dependency-constrained training, building upon the LP approach. Our framework leverages a deep embedding network with self-supervised equidistant regularization loss to project multi-hot label vectors into evenly spaced dense vectors, which not only facilitate problem transformation but also contain latent activity patterns, including label correlations. We treat these dense vectors as new labels and design a corresponding deep learning and classification (DLC) strategy that optimizes all GAR model parameters except for the classification layers. Additionally, we propose a training-only auxiliary branch, dubbed TransOvR module, to bolster the model’s capacity for inter-activity dependency reasoning. By leveraging Transformer’s context-aware capability, this module facilitates interactions between multiple binary classifiers, focusing on correlations between activities. Our extensive experiments demonstrate the effectiveness of our method, outperforming state-of-the-art multi-label GAR methods.}, archive = {J_ASOC}, author = {Lindong Li and Linbo Qing and Pingyu Wang and Yang Xiao and Wang Tang and Yonghong Peng}, doi = {10.1016/j.asoc.2025.113721}, journal = {Applied Soft Computing}, month = {12}, pages = {113721}, shortjournal = {Appl. Soft. Comput.}, title = {Equidistant deep embedding-based multi-label group activity recognition with dependency-constrained training}, volume = {184}, year = {2025}, }
• (2025). Advanced nonlinear optimization of low- and high-resolution medical images using adaptive deep spearman correlation analysis (D-SCA) for pattern sequence recognition. ASOC, 184, 113720. (www)

  @article{ , abstract = {This paper presents the state-of-art model for those healthcare multimedia datasets that have nonlinear limitations during the processing of multi-dimensional or non-linear data that usually decrease the accuracy rate and increase computational cost as per computational resources. Such as correlation among the low-resolution images to high-resolution images, and videos, the reason is non-linearity between data, non-linear dimensions or variation and illusion in the data. The proposed state-of-the-art model uses multiple techniques with transfer learnings approach to solve these issues. This paper further explains the details of the proposed deep convolutional neural network model for visual feature maps extraction and then followed by the proposed algebraic extension of Spearman correlation analysis to create the relationship of low-resolution image to high-resolution image with adaptive optimization. Then it leads to the next component of model as Radial Basis Function Network (RBFN) for non-linear mapping among LR to HR images, for classification the hybrid Xception deep learning model is used and implemented on four benchmark datasets.}, archive = {J_ASOC}, author = {Muhammad Saddam Khokhar and Misbah Ayoub and Zakria and Abdullah Lakhan}, doi = {10.1016/j.asoc.2025.113720}, journal = {Applied Soft Computing}, month = {12}, pages = {113720}, shortjournal = {Appl. Soft. Comput.}, title = {Advanced nonlinear optimization of low- and high-resolution medical images using adaptive deep spearman correlation analysis (D-SCA) for pattern sequence recognition}, volume = {184}, year = {2025}, }
• (2025). Cross-modal deep interaction and modal-aware aggregation network for visible and infrared tracking. ASOC, 184, 113719. (www)

  @article{ , abstract = {Currently, developing a robust RGB-Thermal (RGBT) tracking method in complex environments remains challenging in effectively mining and fusing cross-modal complementary information, enhancing modal perception capabilities, and improving the representation ability of fused semantic features. To address these challenges, we propose a novel architecture called Multi-branch Cross-modal Deep Interaction Fusion and Adaptive Aggregation Integrating Hybrid Attention Semantic Enhancement Network (MCFTNet). Specifically, MCFTNet first extracts modality-specific features through dedicated branches, followed by designing a cross-modal deep interaction fusion network to achieves deep interaction and comprehensive fusion of cross-modal features through a fusion branch as the medium. Furthermore, a modal-aware adaptive aggregation module is developed to dynamically aggregates high-resolution features from different branches, while significantly enhancing the discriminative ability of multi-modal features. Finally, a hybrid attention semantic enhancement module is introduced that combines carefully designed enhanced multi-head attention and hierarchical attention to optimize the fused semantic features, thereby achieving highly accurate prediction of target position and shape. Extensive experiments on three mainstream public benchmark datasets, GTOT, RGBT234, and LasHeR, demonstrate the effectiveness and robustness of our proposed method.}, archive = {J_ASOC}, author = {Xiang Liu and Haiyan Li and Victor Sheng and Yujun Ma and Xiaoguo Liang and Guanbo Wang}, doi = {10.1016/j.asoc.2025.113719}, journal = {Applied Soft Computing}, month = {12}, pages = {113719}, shortjournal = {Appl. Soft. Comput.}, title = {Cross-modal deep interaction and modal-aware aggregation network for visible and infrared tracking}, volume = {184}, year = {2025}, }
• (2025). Embedding neural sampling and adversarial bandit into gene expression programming for symbolic regression. ASOC, 184, 113718. (www)

  @article{ , abstract = {Evolutionary computation methods for symbolic regression problems use a search method to actively explore the mathematical expression space to find a solution. However, the evolutionary computation search process lacks a clear direction due to the randomness of evolutionary operations. In contrast, deep learning methods focus on learning a clear mapping from a given dataset to a mathematical function, but the learning process does not generally involve an active search of the potential solution space. To combine the advantages of the active search of evolutionary computation and the clear mapping direction of neural networks, this paper proposes a novel algorithm called GVAE-ABGEP. GVAE-ABGEP incorporates a grammar variational autoencoder and adversarial bandit into gene expression programming to guide its search process. GVAE-ABGEP partitions the mathematical expression space into many subspaces. It then leverages an adversarial bandit — AvgHExp3 to choose a subspace. In the selected subspace, it then utilizes an autoencoder to sample individuals whose fitnesses are near the local optima, and executes crossover, mutation, and selection in evolutionary computation to explore the global optima. Experiments on 18 symbolic regression and 12 physics benchmarks show that GVAE-ABGEP outperforms three baseline gene expression programming methods and six baseline machine learning methods.}, archive = {J_ASOC}, author = {Qiang Lu and Qiuchen Yuan and Dawei Li and Congwen Xu and Jake Luo and Zhiguang Wang}, doi = {10.1016/j.asoc.2025.113718}, journal = {Applied Soft Computing}, month = {12}, pages = {113718}, shortjournal = {Appl. Soft. Comput.}, title = {Embedding neural sampling and adversarial bandit into gene expression programming for symbolic regression}, volume = {184}, year = {2025}, }
• (2025). Graph-based analysis of problem space and genetic programming classifier performance using optimal transport dataset distance. ASOC, 184, 113716. (www)

  @article{ , abstract = {Understanding and interpreting machine learning tasks is a non-trivial endeavor. To this end, many works have proposed creative ways to organize and visualize a problem dataset, the learning process itself, or the models produced after learning has occurred. Moreover, as the scope of machine learning increases, the re-purposing or specialization of models becomes more common, or as continued automation is sought for algorithm deployment, understanding sets of problems has also become relevant. The present work proposes a graph-based representation of a set of machine learning problems, a novel characterization of the relationships and structures that are present in such a set. While similar proposals have been made before, the proposed methodology employs a recent and unique metric between problem datasets, the Optimal Transport Dataset Distance, which allows for the computation of a mathematically rigorous distance matrix for problem sets. This allows for the construction of a graph-based representation, while previous works relied on representations based on problem meta-features that were defined heuristically. Results show that the resulting graph representation of a problem set exhibits structural properties that are related to empirical indicators of problem difficulty, such as the average error, the size of the dataset, and the class imbalance. A similar analysis using meta-features shows that the structure of the resulting graphs cannot capture the same nuanced relationships between problems.}, archive = {J_ASOC}, author = {Joel Lee Nation and Daniel Fajardo and Yuliana Martínez and Arnoldo Díaz-Ramírez and Leonardo Trujillo}, doi = {10.1016/j.asoc.2025.113716}, journal = {Applied Soft Computing}, month = {12}, pages = {113716}, shortjournal = {Appl. Soft. Comput.}, title = {Graph-based analysis of problem space and genetic programming classifier performance using optimal transport dataset distance}, volume = {184}, year = {2025}, }
• (2025). Graph-based deep reinforcement learning for dynamic scheduling of flexible job-shop considering worker fatigue and multi-skill factors. ASOC, 184, 113712. (www)

  @article{ , abstract = {In recent years, the dynamic scheduling problem of flexible job shops (DFJSP) has garnered significant attention. However, most DFJSP studies focus solely on machine constraints, often overlooking the crucial factor of worker constraints. As an essential resource within the production process, the efficient utilization of workers can substantially enhance production efficiency. Consequently, the scheduling problem in dual-resource constrained (DRC) production systems, where operation times fluctuate according to worker skill levels and fatigue, is investigated. Considering new job arrivals as dynamic events, this study comprehensively integrates worker skill and fatigue factors, with the goal of minimizing total cost and makespan by developing a dual-resource constrained dynamic flexible job-shop scheduling (DRC-DFJSP) optimization model. To enable real-time model resolution in dynamic environments, an end-to-end deep reinforcement learning (DRL) scheduling method is introduced. The decision-making process is formulated as a Markov decision process (MDP) and guided by a reward mechanism tailored to optimization objectives. An enhanced proximal policy optimization (PPO) algorithm, combined with an adaptive clipping mechanism and a prioritized experience replay buffer, is applied to handle operation ordering, machine assignment, and worker allocation within the DRC-DFJSP framework. To further enhance decision-making capabilities, an attention-based graph neural network feature extraction method is incorporated to capture the intricate connections between operations, machines, and workers, resulting in a more precise characterization of the workshop state. Numerical experiments and case studies demonstrate that the proposed scheduling method surpasses existing strategies in dynamic environments that account for worker skill and fatigue levels.}, archive = {J_ASOC}, author = {Yiwen Hu and Zequn Zhang and Jie Chen and Dunbing Tang and Qixiang Cai}, doi = {10.1016/j.asoc.2025.113712}, journal = {Applied Soft Computing}, month = {12}, pages = {113712}, shortjournal = {Appl. Soft. Comput.}, title = {Graph-based deep reinforcement learning for dynamic scheduling of flexible job-shop considering worker fatigue and multi-skill factors}, volume = {184}, year = {2025}, }
• (2025). Command and control network architecture optimization method based on dual-layer weighting and TOPSIS-GRA of indicators. ASOC, 184, 113711. (www)

  @article{ , abstract = {In modern warfare, the effectiveness of command and control (C2) networks plays a pivotal role in shaping operational performance and determining combat outcomes. Even under identical resource allocations, variations in command strategies and network architectures can lead to vastly different results. While many existing optimization methods rely on complex network theory to enhance system robustness, they often exhibit randomness and overlook commanders’ intent and expert knowledge during architecture design. Unlike civilian networks, military C2 networks are highly dependent on operational planning and domain expertise, emphasizing decision transparency and interpretability. Subjective evaluation methods-such as those based on rules or expert logic-are more intuitive for commanders but suffer from bias and inconsistency. Conversely, purely data-driven models like deep learning lack transparency, limiting their practical utility in high-stakes scenarios. To bridge this gap, this study proposes a dual-layer weighting optimization approach based on the TOPSIS and Grey Relational Analysis (GRA) methods. First, C2 network architectures are constructed using complex network theory. Then, an evaluation index system is developed around four key metrics: invulnerability, communication efficiency, connectivity, and network density. Subjective weights are derived via the Analytic Hierarchy Process (AHP), while objective weights are calculated using an improved entropy method. These are integrated into a variable-weight model to obtain final weights for optimization. By combining TOPSIS and GRA, the method accounts for both subjective preferences and objective measurements, ensuring a balanced evaluation. Simulation results demonstrate that the proposed method comprehensively considers both subjective and objective factors as well as the impact of state changes on weight adjustments, aligning more closely with real battlefield environments. The constructed C2 network architecture can adapt to different combat missions, guided by subjective evaluations of commander intent and expert experience while being refined through objective assessments to achieve adaptive adjustments. The ranking results account for both the distance and shape variations among indicators. Statistical analysis and comparisons with other methods further validate the effectiveness of the proposed approach, which also reduces computation time by a factor of five compared to alternative algorithms, highlighting its superiority.}, archive = {J_ASOC}, author = {Jianwei Wang and Qing Zhang and Chengsheng Pan}, doi = {10.1016/j.asoc.2025.113711}, journal = {Applied Soft Computing}, month = {12}, pages = {113711}, shortjournal = {Appl. Soft. Comput.}, title = {Command and control network architecture optimization method based on dual-layer weighting and TOPSIS-GRA of indicators}, volume = {184}, year = {2025}, }
• (2025). A distributed learning framework with blockchain and privacy-preserving for IoV. ASOC, 184, 113710. (www)

  @article{ , abstract = {The Internet of Vehicles (IoV), as a critical component of the Internet of Things (IoT), constructs a distributed system comprising vehicles, roadside units (RSUs), and cloud servers. In the IoV environment, the secure sharing of data and the protection of privacy are of paramount importance, as they directly impact the decision-making processes of intelligent vehicles and overall road safety. Given the openness of IoV, it faces risks of privacy leakage and poisoning attacks during data exchange and model training, which threaten the integrity and reliability of the data. To address these challenges, this study proposes a privacy protection framework that integrates blockchain technology and differential privacy. This framework incorporates dual differential privacy techniques within federated learning to enhance data privacy protection and designs a dynamic gradient aggregation mechanism to defend against data poisoning attacks, thereby ensuring the security of the data. Experimental results demonstrate that this framework maintains high model accuracy even under attack rates of up to 30%, exhibiting remarkable resilience against such attacks. Overall, this study emphasizes the significance of data security and privacy protection in the IoV domain and illustrates the potential of blockchain and differential privacy technologies in enhancing the security of IoV data and safeguarding user privacy. This research provides robust support for the sustainable development of IoV.}, archive = {J_ASOC}, author = {Chunhai Li and Yan Long and Yong Ding and Changsong Yang and Chuan Zhang and Meng Shen and Liehuang Zhu}, doi = {10.1016/j.asoc.2025.113710}, journal = {Applied Soft Computing}, month = {12}, pages = {113710}, shortjournal = {Appl. Soft. Comput.}, title = {A distributed learning framework with blockchain and privacy-preserving for IoV}, volume = {184}, year = {2025}, }
• (2025). Chaotic vortex nanoparticle swarms (CVNPS): Formation and stability analysis. ASOC, 184, 113708. (www)

  @article{ , abstract = {Vortex Nanoparticle Swarms (VNPS) are a class of self-organizing systems that exhibit vortex-like collective behavior but often suffer from dynamic instability and poor angular momentum conservation. While VNPS may achieve partial spatial stability, coherent motion is typically unsustained. To address this, the Chaotic Vortex Nanoparticle Swarm (CVNPS) model is proposed, incorporating chaotic perturbations via logistic maps within a discrete particle model governed by the Generalized Morse Potential. This integration enhances system stability by accelerating convergence, reducing energy fluctuations, and promoting sustained vortex formation. Comparative analysis using key performance indicators – average nearest neighbor distance, polarization, angular momentum, swarm diameter, total energy, energy dissipation rate, and variance – demonstrates that CVNPS significantly outperforms traditional VNPS. It achieves faster stabilization and maintains coherent structure over time, even under varied initial conditions. By addressing both configurational and dynamical stability, CVNPS establishes a foundation for more reliable nanoparticle swarm applications in nanotechnology.}, archive = {J_ASOC}, author = {Mahvish Khurshid Bijli and Prabal Verma and Amrit Pal Singh}, doi = {10.1016/j.asoc.2025.113708}, journal = {Applied Soft Computing}, month = {12}, pages = {113708}, shortjournal = {Appl. Soft. Comput.}, title = {Chaotic vortex nanoparticle swarms (CVNPS): Formation and stability analysis}, volume = {184}, year = {2025}, }
• (2025). Aspect-based context iterative network with hyperbolic tangent function for aspect-based sentiment classification. ASOC, 184, 113707. (www)

  @article{ , abstract = {In the domain of natural language processing, sentiment analysis emerges as a pivotal undertaking. Especially, aspect-based sentiment classification has garnered significant attention in recent years, as it can identify and evaluate emotions related to specific aspects within sentences. Existing approaches typically achieve satisfactory results by extracting keyword information from the context to identify polarity. However, a common challenge faced by these methods is the inclusion of irrelevant words in the extracted keywords, leading to decreased classification accuracy. To tackle this issue, we propose an aspect-based context multiple iteration approach, which leverages the correlation between aspects and context and employs the multi-head attention mechanism to iteratively extract contextual keywords. By doing so, we aim to mitigate the interference of irrelevant words and enhance the accuracy of sentiment classification. Additionally, we address the peculiarities of hard samples by introducing a novel loss function that cleverly incorporates the hyperbolic tangent function and allows for improved model accuracy. To validate the effectiveness of our proposed approach, we conduct extensive experiments on four widely datasets and demonstrate the efficacy of our model in improving sentiment classification.}, archive = {J_ASOC}, author = {Chao Zhu and Benshun Yi and Laigan Luo}, doi = {10.1016/j.asoc.2025.113707}, journal = {Applied Soft Computing}, month = {12}, pages = {113707}, shortjournal = {Appl. Soft. Comput.}, title = {Aspect-based context iterative network with hyperbolic tangent function for aspect-based sentiment classification}, volume = {184}, year = {2025}, }
• (2025). MOGA: Multi-objective genetic algorithm to select stacking ensemble learning for classification. ASOC, 184, 113706. (www)

  @article{ , abstract = {Stacking Ensemble Learning (SEL) has been effectively integrated with Multi-Objective Optimisation (MOO) heuristics for classification tasks across various domains, including finance, healthcare, and cybersecurity. This study aims to address the challenge of generalising SEL to a diverse set of classification cases. Thus, the Multi-Objective Genetic Algorithm (MOGA) framework is proposed, utilising a Genetic Algorithm (GA) to evolve a population of distinct SELs, each built from a varied subset of base models. The goal is to select the subset that composes the most effective SEL for a given classification task. MOGA is designed with two main objectives—maximising p r e c i s i o n and r e c a l l —which helps to maintain independence from any specific classification case. In addition, incorporating models of varied types ensures adaptability and high performance in different situations. Comprehensive experimentation was conducted on 23 diverse datasets, where MOGA demonstrated high performance in nearly all datasets, outperforming other ensemble learning (EL) methods in 100% of the datasets in p r e c i s i o n , 78% in r e c a l l , 69.5% in f 1 − s c o r e , and 78% in a c c u r a c y . A t-test analysis yielded results of p -value < 0 . 05 , indicating a statistically significant improvement in the a c c u r a c y of the MOGA over the base models. Moreover, the framework’s application can be extended to regression tasks as well.}, archive = {J_ASOC}, author = {Abdellah Rezoug and Mohamed Bader-el-den}, doi = {10.1016/j.asoc.2025.113706}, journal = {Applied Soft Computing}, month = {12}, pages = {113706}, shortjournal = {Appl. Soft. Comput.}, title = {MOGA: Multi-objective genetic algorithm to select stacking ensemble learning for classification}, volume = {184}, year = {2025}, }
• (2025). Deep reinforcement learning with evolutionary algorithm-guided imitation for capacitated vehicle routing problems. ASOC, 184, 113705. (www)

  @article{ , abstract = {The capacitated vehicle routing problem (CVRP) is a complex combinatorial optimization challenge that seeks to determine cost-effective routes for customer deliveries while adhering to specific capacity constraints. Although deep reinforcement learning (DRL) has shown promise in addressing CVRP, it often encounters issues such as slow convergence and suboptimal accuracy. This study introduces an innovative approach that enhances both convergence efficiency and solution quality by integrating DRL with imitation learning (IL), utilizing an evolutionary algorithm (EA) as an expert. The proposed methodology incorporates an attention mechanism-based neural network to effectively capture the intricate features of CVRP. It leverages IL to use EA-generated solutions as expert demonstrations, thereby guiding the DRL model toward a more efficient exploration of the solution space. The REINFORCE algorithm with baseline is employed to ensure stable and rapid training of the DRL model. Experimental results indicate that this hybrid approach significantly outperforms widely adopted baseline methods and approaches the performance levels of advanced algorithms like LKH3. Furthermore, the method demonstrates robust generalization capabilities across various CVRP instances, underscoring its potential for practical applications in diverse routing scenarios. This research contributes to the field by demonstrating how integrating EA as experts within an IL framework can effectively enhance DRL for solving CVRP.}, archive = {J_ASOC}, author = {Wenqiang Zhang and Xiaomeng Wang and Yashuan Mu and Miaolei Deng and Peng Li}, doi = {10.1016/j.asoc.2025.113705}, journal = {Applied Soft Computing}, month = {12}, pages = {113705}, shortjournal = {Appl. Soft. Comput.}, title = {Deep reinforcement learning with evolutionary algorithm-guided imitation for capacitated vehicle routing problems}, volume = {184}, year = {2025}, }
• (2025). A focal quotient gradient system method for deep neural network training. ASOC, 184, 113704. (www)

  @article{ , abstract = {This paper introduces a novel approach for training deep neural networks, leveraging a mini-batch focal quotient gradient system (QGS-Focal). The proposed method addresses critical challenges in imbalanced datasets and optimization efficiency. By introducing residual constraints into the loss function, we construct a quotient gradient system that effectively mitigates model overfitting and gradient vanishing problems. By incorporating a focal loss mechanism, it innovatively solves data imbalance issues at the algorithmic level. The adoption of a mini-batch strategy and limited memory method significantly reduces computational costs. Our comprehensive experiments on imbalanced CIFAR-10 and CIFAR-100 have demonstrated the superiority of QGS-Focal, achieving 83.4 % precision, 83.6 % recall, and 82.9 % F1-score on CIFAR-10, significantly outperforming SGD, Adam, and QGS. Moreover, our approach reduces training time by 13.3 %, enhancing computational efficiency while maintaining superior classification performance. The t-SNE visualization further confirms that QGS-Focal has superior convergence properties compared to traditional optimization approaches.}, archive = {J_ASOC}, author = {Caili Lv and Xian-long Lv and Zhiyuan Wang and Tianqi Zhao and Wei Tian and Qingqing Zhou and Lin Zeng and Min Wan and Chenghu Liu}, doi = {10.1016/j.asoc.2025.113704}, journal = {Applied Soft Computing}, month = {12}, pages = {113704}, shortjournal = {Appl. Soft. Comput.}, title = {A focal quotient gradient system method for deep neural network training}, volume = {184}, year = {2025}, }
• (2025). Partial differential equations and machine learning integration for transit-oriented development. ASOC, 184, 113703. (www)

  @article{ , abstract = {The accurate classification of rail transit stations is critical for advancing Transit-Oriented Development (TOD) and promoting sustainable urban growth. This research presents a novel hybrid framework that integrates Partial Differential Equations (PDEs) with Machine Learning (ML) techniques for the classification of rail transit stations. Unlike conventional TOD models, this study applies the heat equation to the Node, Place, Ridership-Time (NPRT) framework, offering a mathematically grounded approach to capture spatial-temporal dynamics in transit systems. This integration represents the first known application of PDE-based physical modeling combined with supervised learning for classifying transit stations within a TOD framework. This approach significantly enhances the model’s interpretability while maintaining competitive prediction accuracy. Through extensive case studies and empirical validation, the PDE-NPRT model demonstrates strong performance, with Mean Squared Error (MSE) values ranging from 0.0075 to 0.0222. Although slightly outperformed by enhanced ML models—such as K-Nearest Neighbors (KNN), Deep KNN (DKNN), and Deep Distributed Neural Networks (DDNN)—which achieve MSEs as low as 0.0034, the PDE-NPRT model offers a more interpretable and theoretically robust alternative. Additionally, the study introduces a multi-layer modeling strategy that combines regression analysis, clustering algorithms, PDEs, and neural networks, further enriching the understanding of ridership patterns and congestion mechanisms. Clustering outcomes are validated through external indices, confirming the alignment of model predictions with real-world site characteristics. This work represents a significant advancement in TOD modeling, offering a robust and explainable tool for urban planners and decision-makers. By bridging advanced mathematical modeling with machine learning, it paves the way for more intelligent, data-driven, and sustainable urban mobility strategies.}, archive = {J_ASOC}, author = {Ahad Amini Pishro and Shiquan Zhang and Alain L’Hostis and Qixiao Hu and Yuetong Liu and Zhengrui Zhang and Van Duc Nguyen and Yongguo Fu and Tianzeng Li}, doi = {10.1016/j.asoc.2025.113703}, journal = {Applied Soft Computing}, month = {12}, pages = {113703}, shortjournal = {Appl. Soft. Comput.}, title = {Partial differential equations and machine learning integration for transit-oriented development}, volume = {184}, year = {2025}, }
• (2025). A dynamic SLAM algorithm based on improved YOLOv9S. ASOC, 184, 113700. (www)

  @article{ , abstract = {Due to the limitations of conventional visual SLAM algorithms and the time-intensive process of semantic segmentation in dynamic environments, this paper proposes a SLAM algorithm that incorporates an enhanced version of YOLOv9S specifically designed for dynamic scenes. The ECASPPELAN and DRepNCSPELAN4 modules have been developed to improve the model’s detection capabilities in dynamic scenes. Secondly, the improved YOLOv9S is integrated into the ORB-SLAM2 framework, which employs bipolar geometry to reject dynamic targets and utilizes a depth separation algorithm to prevent the erroneous rejection of static features. Subsequently, the detection performance of the model was evaluated on the PASCAL VOC dataset, resulting in a 2% improvement in mAP_0.5 and a 1.3% improvement in mAP_0.5:0.95. The tracking performance of the system is examined in the TUM dynamic dataset, where the absolute trajectory root mean square error (RMSE) is reduced by 97.5%, and the relative rotational RMSE is reduced by 92.9% in comparison to ORB-SLAM2 in the high-dynamics scenario. Additionally, the average running rate surpasses 30 Hz while constructing high-quality dense maps in real indoor dynamic environments. This allows the robot to receive more detailed texture information. The experimental results demonstrate that the enhanced algorithm presented in this paper exhibits enhanced accuracy and resilience in dynamic scenarios.}, archive = {J_ASOC}, author = {Qiguang Zhu and Yuchao Zhao and Haofeng Zhang and Weidong Chen}, doi = {10.1016/j.asoc.2025.113700}, journal = {Applied Soft Computing}, month = {12}, pages = {113700}, shortjournal = {Appl. Soft. Comput.}, title = {A dynamic SLAM algorithm based on improved YOLOv9S}, volume = {184}, year = {2025}, }
• (2025). Neural semantic evaluation of blockchain policy tools in china. ASOC, 184, 113699. (www)

  @article{ , abstract = {To explore the differences in policies guiding the development of the blockchain industry in different regions of mainland China, in this study, we developed a compound neural network model comprising bidirectional long short term memory and deep biaffine attention models that analyses the semantic texts of blockchain policies issued by 31 provinces in mainland China. Machine learning models — specifically, term frequency–inverse document frequency and K-means models are used to implement feature selection of the policy text matrix classification after semantic analysis. Finally, this study proposes an innovative policy tools matching approach. We construct a word-topological map for each text category based on semantic relationships. To validate the effectiveness of these tools, we conduct an empirical analysis using a multivariate linear regression model. The results demonstrate that blockchain policy tools significantly promote blockchain innovation in Mainland China.}, archive = {J_ASOC}, author = {Yuxi Zhang and Haifeng Guo and Ke Peng and Hongzhi Wang}, doi = {10.1016/j.asoc.2025.113699}, journal = {Applied Soft Computing}, month = {12}, pages = {113699}, shortjournal = {Appl. Soft. Comput.}, title = {Neural semantic evaluation of blockchain policy tools in china}, volume = {184}, year = {2025}, }
• (2025). Explainable nature-inspired optimization via virtual and actual multi-objective strategies to establish a smart earthquake early warning system. ASOC, 184, 113698. (www)

  @article{ , abstract = {Geosynthetic-reinforced soil (GRS) structures are considered for reducing displacement and providing economical reinforcement solutions. The risk assessment of these structures against earthquakes, based on the prediction of seismic sliding displacement, is a major challenge in this field. Multi-objective optimization is a powerful machine learning tool for selecting efficient features for high-performance forecasting. This research investigates two strategies based on swarm intelligence and genetic programming for a comprehensive evaluation. These frameworks integrate multiobjective optimization algorithms and Newmark methods for utilizing effective physics-informed features. The first strategy is virtual multi-objective (VMO) optimization by applying particle swarm optimization (PSO) based on minimizing one function via variations of other functions. In this approach, the error function, as a computational error object, is minimized versus the nomination of interpretable feature space as a computational cost object through the virtual Pareto front. The second strategy is actual multi-objective (AMO) optimization by exploiting nondominated sorting genetic algorithm II (NSGA-II) based on minimizing several functions simultaneously with two various approaches, including bi-objective and many-objective algorithms through actual Pareto-optimal solutions. In this approach, the computational error value, computational cost value, and computational time value are minimized at the same time. The main novelty of the first technique is low computational complexity, resulting in high speed due to definite search space dimension-based exploration and exploitation to forecast seismic sliding displacement, whereas the major achievement of the second technique is high computational accuracy due to multiobjective structure-assisted exploitation and exploitation. Through numerical validation by employing the Newmark methods, the resultant model predicts the seismic sliding displacement of these structures using two algorithms efficiently. Nevertheless, both strategies have good performance for intelligent forecasting. The actual many-objective optimization algorithm is a more effective switchable machine learning tool based on the proposed adaptable performance index for developing a smart earthquake early warning software that can precisely detect imminent natural hazards.}, archive = {J_ASOC}, author = {Milad Zarchi and Reza A. Nazari and Kong Fah Tee}, doi = {10.1016/j.asoc.2025.113698}, journal = {Applied Soft Computing}, month = {12}, pages = {113698}, shortjournal = {Appl. Soft. Comput.}, title = {Explainable nature-inspired optimization via virtual and actual multi-objective strategies to establish a smart earthquake early warning system}, volume = {184}, year = {2025}, }
• (2025). DTKD-DL: Dual-teacher knowledge distillation with dual-loops for continuous few-shot relation extraction. ASOC, 184, 113696. (www)

  @article{ , abstract = {This paper introduces a new model named DTKD-DL, designed to address the issue of Continuous Few-Shot Relation Extraction (CFRE) across tasks, with the goal of learning and adapting to newly emerging relations while reducing catastrophic forgetting. In this paper, we have designed a dual-teacher knowledge distillation model based on relation information to enrich knowledge representation and retain prior knowledge. We employ a dual-loops distillation approach, which facilitates knowledge transfer and optimizes the direction of parameter updates, thereby reducing the occurrence of catastrophic forgetting. Furthermore, to avoid overfitting issues caused by multiple rounds of distillation, we have innovatively integrated reinforcement learning with the model. We have validated our model on the FewRel and TACRED datasets and compared it with the large language model Llama3-8b, demonstrating the effectiveness of our model in this scenario and its advantages over the most advanced methods, achieving state-of-the-art results.}, archive = {J_ASOC}, author = {Ruifeng Xu and Yi Chen and Zhongyan Yi and Shun Huang and Liang He}, doi = {10.1016/j.asoc.2025.113696}, journal = {Applied Soft Computing}, month = {12}, pages = {113696}, shortjournal = {Appl. Soft. Comput.}, title = {DTKD-DL: Dual-teacher knowledge distillation with dual-loops for continuous few-shot relation extraction}, volume = {184}, year = {2025}, }
• (2025). A two-stage feature selection approach with fuzzy covering-based rough sets based on discernibility matrix. ASOC, 184, 113695. (www)

  @article{ , abstract = {Fuzzy β -covering has attracted considerable research attention due to its enhanced capability to accurately and comprehensively represent uncertain information. Unlike partition-based approaches, fuzzy β -covering maintains its covering properties when augmented with additional elements. This inherent flexibility necessitates rigorous redundancy analysis. Hence, the calculation of the reduct of fuzzy β -covering relative to decision attribute constitutes a fundamental challenge in this context. To address this issue, we formally define the concepts of indispensable covering elements and core of fuzzy β -covering with respect to the decision attribute. Then, to facilitate efficient computation of core and reduct, we introduce the discernibility matrix for fuzzy β -covering and establish equivalent representations of the core and reduct. Furthermore, we conduct systematic examinations to verify the indispensability of some existing concepts in fuzzy β -covering. Meanwhile, a significant limitation of current feature selection methods useing fuzzy β -covering lies in their computational complexity, primarily due to repeated recalculation of dependency functions during iterations. To overcome this limitation, we propose an efficient feature selection algorithm that identifies all reducts directly through the discernibility matrix. Furthermore, we propose a two-stage feature selection method for fuzzy β -covering, which iteratively eliminates redundant features to identify optimal feature subsets in each iteration. Finally, we verify through time complexity analysis and numerical experiments that the proposed algorithm significantly outperforms existing feature selection algorithms in computational efficiency. Comparative analysis further reveals that our method achieves superior results in terms of selected feature subset sizes and classification accuracy.}, archive = {J_ASOC}, author = {Tianyu Wang and Shuai Liu and Bin Yang}, doi = {10.1016/j.asoc.2025.113695}, journal = {Applied Soft Computing}, month = {12}, pages = {113695}, shortjournal = {Appl. Soft. Comput.}, title = {A two-stage feature selection approach with fuzzy covering-based rough sets based on discernibility matrix}, volume = {184}, year = {2025}, }
• (2025). A multi-source remote sensing image watermarking model based on multi-domain generative adversarial networks. ASOC, 184, 113694. (www)

  @article{ , abstract = {Due to the heterogeneity stemming from distinct sensor types, quantized spatial resolution levels, and categorized acquisition angles within multi-source remote sensing images, their data distribution and characteristics vary significantly, leading to poor cross-domain consistency and weak resistance to domain-specific distortions in traditional watermarking approaches. Traditional watermarking methods often struggle to handle such heterogeneous datasets. To address this challenge, this paper presents a watermarking model based on a multi-domain generative adversarial network (MD-GAN). By incorporating a multi-domain adversarial training mechanism and designing multiple domain discriminators, the generator is optimized to work across various image domains, thereby enhancing model robustness. Additionally, a multi-scale generator is utilized to account for differences in spatial resolution, improving both the watermark’s concealment and extraction accuracy. Experimental results demonstrate that the MD-GAN model achieves a 7 % improvement in robustness over the state-of-the-art models in terms of resistance to noise attacks and geometric transformations. Ablation studies confirm that the multi-domain adversarial training mechanism and multi-scale generator contribute significantly to this enhancement. MD-GAN provides a powerful solution for the copyright protection of multi-source remote sensing images, with substantial potential for real-world applications such as secure transactions and data sharing.}, archive = {J_ASOC}, author = {Heyan Wang and Xingxiang Jiang and Minxuan Wang and Changqing Zhu and Na Ren and Luanyun Hu}, doi = {10.1016/j.asoc.2025.113694}, journal = {Applied Soft Computing}, month = {12}, pages = {113694}, shortjournal = {Appl. Soft. Comput.}, title = {A multi-source remote sensing image watermarking model based on multi-domain generative adversarial networks}, volume = {184}, year = {2025}, }
• (2025). Multiscale convolution-attention model for efficient alzheimer’s disease and mild cognitive impairment diagnosis approach using sMRI. ASOC, 184, 113693. (www)

  @article{ , abstract = {Alzheimer's disease (AD) affects over 50 million people worldwide and causes a gradual decline in memory, language, and actions, with no known cure. Early identification and treatment are crucial to improve the quality of life of affected individuals. Deep learning algorithms based on brain structural magnetic resonance imaging (sMRI) show promise in predicting AD. However, relying solely on deep convolutional neural network (CNN) architecture has limitations in global modeling. Transformer, which is a feature learning technique, has excelled in computer vision applications, sparking interest in its application in brain image processing. However, pure Transformer architectures encounter challenges when trained on small sMRI datasets. Meanwhile, CNN-based methods do not consider the interdependence between voxels, which hinders the comprehensive understanding of the global characteristics of sMRI data. To address these challenges, an effecient deep learning framework was developed that combines the benefits of both the CNN and Transformer. In the preliminary phase of feature extraction, we introduce a multiscale feature fusion stem that employs convolutional kernels of varying scales to derive low-level features from the sMRI and integrate them to enhance the recognition accuracy. Furthermore, the proposed method introduces convolutional split attention with a squeeze and excitation block and additional convolution operations in the core sections of the Transformer, thereby enabling multiscale feature extraction and fusion. The model integrates multi-head light self-attention and a sandglass local feed-forward network block for classifier modeling, enhancing the extraction of sMRI features with a MobileNet cost. The proposed model, which was tested on the Alzheimer's Disease Neuroimaging Initiative dataset, achieved superior performance in AD diagnosis with fewer parameters and reduced computational costs, demonstrating its potential as a state-of-the-art solution.}, archive = {J_ASOC}, author = {Uttam Khatri and Jun-Hyung Kim and Goo-Rak Kwon}, doi = {10.1016/j.asoc.2025.113693}, journal = {Applied Soft Computing}, month = {12}, pages = {113693}, shortjournal = {Appl. Soft. Comput.}, title = {Multiscale convolution-attention model for efficient alzheimer’s disease and mild cognitive impairment diagnosis approach using sMRI}, volume = {184}, year = {2025}, }
• (2025). F2CAU-net: A dual fuzzy medical image segmentation cascade method based on fuzzy feature learning. ASOC, 184, 113692. (www)

  @article{ , abstract = {Accurate medical image segmentation plays a pivotal role in clinical diagnosis and treatment planning. However, existing methods – particularly those based on U-Net architectures – still face considerable difficulties when dealing with ambiguous boundaries, low-contrast regions, and noise artifacts. These challenges arise from the deterministic nature of conventional convolutional and attention mechanisms, which are often inadequate in modeling the inherent uncertainty and variability present in medical images. To tackle these limitations, we propose F 2 CAU-Net, a Dual Fuzzy Medical Image Segmentation Cascade Method that integrates a fuzzy convolution module for better modeling of uncertain and imprecise features, and a fuzzy attention mechanism to suppress redundancy and enhance focus on regions of interest. The proposed method captures both local and global contextual fuzzy information to improve segmentation accuracy and robustness. Extensive experiments on multiple benchmark datasets – including COVID-19 lesions, brain tumors, and skin cancer – demonstrate that F 2 CAU-Net significantly outperforms existing state-of-the-art models, particularly in scenarios with complex boundaries and noise. This approach offers promising potential for clinical applications, providing a more reliable and uncertainty-aware solution for medical image analysis.}, archive = {J_ASOC}, author = {Tianyi Zhou and Haipeng Wang and Sheng Geng and Hengrong Ju and Jiashuang Huang and Fan Fu and Weiping Ding}, doi = {10.1016/j.asoc.2025.113692}, journal = {Applied Soft Computing}, month = {12}, pages = {113692}, shortjournal = {Appl. Soft. Comput.}, title = {F2CAU-net: A dual fuzzy medical image segmentation cascade method based on fuzzy feature learning}, volume = {184}, year = {2025}, }
• (2025). A physics-informed mitigation method for DC microgrids under cyber attacks. ASOC, 184, 113691. (www)

  @article{ , abstract = {Cyber attacks pose serious threats to DC microgrids, making effective mitigation strategies very critical. However, most existing mitigation schemes ignore the physical characteristics of microgrids, which may lead to inaccurate attack detection and affect the mitigation performance. To address this issue, a physics-informed mitigation method for cyber attacks on DC microgrids is proposed. A cyber–physical framework of the DC microgrid is established, and the impact of cyber attacks on the microgrid is analyzed. The denoising autoencoder is utilized to improve the quality of the input data, and then the consistency characteristic of the DC microgrid is incorporated as a physical-informed constraint into the training of the state estimation model. Subsequently, anomaly detection is performed by comparing the estimated state and the real-time measured state. Once the attacks are detected, the estimated state is used to compensate the control inputs, mitigating the effect of attacks and ensuring the safe and stable operation of the DC microgrid. Simulation results show that compared with existing neural network-based estimators, the proposed physics-informed state estimation model can enhance the accuracy by 7 . 98 % ∼ 12 . 54 % , providing more precise estimated states to mitigate the effects of attacks.}, archive = {J_ASOC}, author = {Wanwan Ren and Jun Peng and Yun Zhou and Weirong Liu and Fu Jiang}, doi = {10.1016/j.asoc.2025.113691}, journal = {Applied Soft Computing}, month = {12}, pages = {113691}, shortjournal = {Appl. Soft. Comput.}, title = {A physics-informed mitigation method for DC microgrids under cyber attacks}, volume = {184}, year = {2025}, }
• (2025). Acquisition of representative data sets by filtering out redundant objects and attributes with fuzzy preference-based rough sets and dominance principles. ASOC, 184, 113690. (www)

  @article{ , abstract = {The dominance principle is crucial for evaluating consistency in dominance-based rough sets, yet redundant objects or attributes impair decision consistency. While existing work focuses on modeling and attribute reduction, object-induced inconsistency remains understudied. This study proposes a novel methodology for acquiring compact datasets through representative object extraction and attribute reduction in fuzzy preference-based rough sets, with a specific focus on preserving consistency in dominance principles. Firstly, the extent of dominance relations is quantified by fuzzy preference relations, while the evaluation of consistency between conditional and decision attributes is accomplished through distance measures. Subsequently, representative objects are identified by eliminating those with lower consistency in dominance principles, as assessed by distance measures with a predefined parameter. Further, a streamlined dataset is achieved through attribute reductions in fuzzy preference-based rough sets, incorporating representative objects. In conclusion, our proposed method is validated using numerical datasets, and its effectiveness is evaluated through measures rooted in rough set theory, machine learning, and statistics. This research contributes to a more comprehensive understanding of dominance-based rough sets by addressing the often-overlooked issue of inconsistency resulting from redundant objects.}, archive = {J_ASOC}, author = {Shuyun Yang and Guang Shi and Yuchao Li}, doi = {10.1016/j.asoc.2025.113690}, journal = {Applied Soft Computing}, month = {12}, pages = {113690}, shortjournal = {Appl. Soft. Comput.}, title = {Acquisition of representative data sets by filtering out redundant objects and attributes with fuzzy preference-based rough sets and dominance principles}, volume = {184}, year = {2025}, }
• (2025). Spiking neural network with time-varying weights for rail squat detection. ASOC, 184, 113689. (www)

  @article{ , abstract = {Axle box acceleration (ABA) measurements can be used for continuously monitoring rail infrastructure and detecting rail surface defects such as squats. However, accurately detecting squats is challenging due to their short-duration responses and low occurrence in ABA signals, particularly for light squats that exhibit subtle ABA responses. To address this challenge, we propose using a spiking neural network (SNN) with time-varying weights to enhance the detection performance of rail squats based on ABA measurements. Our approach employs a simple SNN architecture without hidden layers, trained using a method that combines genetic algorithms, k-fold cross-validation, and multi-start gradient-based approach to optimise hyperparameters and weights. The proposed methodology demonstrates competitive accuracy compared to other state-of-the-art SNN-based methods on UCI benchmarks for both binary and multi-class nonlinear problems. Part of its advantages include higher efficiency with a simpler architecture and training approach that reduces computational times while achieving effective spatiotemporal pattern detection. As shown by real-field measurements from Dutch and Swedish railways in anomaly detection, it effectively captures subtle changes in light squat defect responses in ABA signals and achieves a detection performance of 100% for severe squat defects and over 93% for light squat defects. Furthermore, we show that the spike responses, postsynaptic potentials, and membrane potentials can be used as a new way to explain and analyse the ABA signals. The proposed method using time-varying weights highlights a correspondence with the physical problem and offers an ability to capture sudden and subtle changes in the responses, which is crucial, particularly for detecting defects in their early stages.}, archive = {J_ASOC}, author = {Wassamon Phusakulkajorn and Jurjen Hendriks and Zili Li and Alfredo Núñez}, doi = {10.1016/j.asoc.2025.113689}, journal = {Applied Soft Computing}, month = {12}, pages = {113689}, shortjournal = {Appl. Soft. Comput.}, title = {Spiking neural network with time-varying weights for rail squat detection}, volume = {184}, year = {2025}, }
• (2025). Spectral–spatial representation progressive learning via segmented attention for 3D skeleton-based motion prediction. ASOC, 184, 113688. (www)

  @article{ , abstract = {Recently, GCNs-based methods have demonstrated impressive performance in human behavior prediction tasks. We believe that human motion modeling can explained as motion correlation extraction from the combination of the active and static motion parts analysis. However, existing methodologies fail to address the issue that feature information associated with static regions may overshadow feature information from dynamic regions, ultimately affecting the extraction of network features. Moreover, the unique low-pass feature pre-retention processing mechanism of GCN on the spectrum will lead to the attitude of some sequences remaining unchanged during the prediction process and further hurt the prediction. In this paper, we propose a Spectral–Spatial Representation Progressive Learning network to solve the problem above. Firstly, we propose a segmented attention block to compare the input observation sequence with the static contrast standard to obtain the motion region and the rest region. Then, we design the Spectrum Deconstruction Recombination Factor block(SDRF) to extract the global bandpass spectrum of human bone joints. The joint features of different regions are encoded by graph convolution and high-frequency feature filter coding based on geometric algebra. Specifically, a spectral–spatial interaction block is presented in each SDRF, focusing on the diversity of motion sequence frequency domain and spatial domain map, and realizes the fine extraction of historical pose sequence features from the two levels of space and spectral domain. Experimental results demonstrate that our approach outperforms state-of-art algorithms by 2.4%, 5.3% and 4.7% in terms of 3D mean per joint position error on Human 3.6M, CMU Mocap and 3DPW datasets, respectively.}, archive = {J_ASOC}, author = {Wenming Cao and Jianhua Zhang and Jianqi Zhong}, doi = {10.1016/j.asoc.2025.113688}, journal = {Applied Soft Computing}, month = {12}, pages = {113688}, shortjournal = {Appl. Soft. Comput.}, title = {Spectral–spatial representation progressive learning via segmented attention for 3D skeleton-based motion prediction}, volume = {184}, year = {2025}, }
• (2025). Multi-modal multi-trip supply chain model aided by smart contract victim tracking — An innovative pathway to disaster management under uncertainty. ASOC, 184, 113687. (www)

  @article{ , abstract = {This study addresses the complex challenges of post-earthquake rescue operations, using the recent earthquake in Turkey as a reference case. It focuses on developing an optimized model for scenarios with acute vehicle shortages, aiming to minimize both operational costs and response time during the critical initial phase of disaster relief. The proposed solution is built upon a robust mathematical framework that employs aerial vehicles for post-disaster area assessment, resource allocation, and the relocation of critically injured victims. The model leverages a soft computing approach, integrating the Weighted Sum Method (WSM) and Neutrosophic Compromise Programming (NCP). To enhance decision-making under uncertainty, the framework incorporates hexagonal type-2 fuzzy defuzzification, a technique grounded in soft computing principles. Results demonstrate the effectiveness of this approach: the NCP method achieved a response time of 213 min (3.55 h) and a cost of Rs 821,026.5, compared to 217.5 min (3.62 h) and Rs 820,860.3 for the WSM method—both successfully coordinating the rescue of 1,450 victims through efficient deployment of drones and helicopters. In addition, the study introduces a decentralized Ethereum-based smart contract to securely store and retrieve critical victim information. Validated through rigorous unit testing, the contract ensures data transparency and integrity, executing at a cost of 0.00379246 Ether. This blockchain-enabled feature complements the core optimization model, supporting real-time, tamper-proof data handling. To further validate the model’s applicability, a second real-life numerical example based on the recent Sikkim cloudburst is analyzed. The findings reinforce the model’s adaptability and practical value. The managerial implications of this research highlight the importance of soft computing-driven decision support, proactive contingency planning, and the integration of intelligent technologies in disaster response. This holistic framework — combining soft computing methodologies, advanced optimization models, and blockchain technology — offers an innovative and scalable solution for enhancing the resilience and efficiency of disaster management supply chains.}, archive = {J_ASOC}, author = {Alisha Roushan and Amrit Das and Anirban Dutta and Uttam Kumar Bera}, doi = {10.1016/j.asoc.2025.113687}, journal = {Applied Soft Computing}, month = {12}, pages = {113687}, shortjournal = {Appl. Soft. Comput.}, title = {Multi-modal multi-trip supply chain model aided by smart contract victim tracking — An innovative pathway to disaster management under uncertainty}, volume = {184}, year = {2025}, }
• (2025). Contrastive learning with hard negatives for sentence embeddings. ASOC, 184, 113685. (www)

  @article{ , abstract = {Unsupervised sentence representation learning remains a core challenge in natural language processing. Recent contrastive learning methods have shown strong potential in capturing sentence-level semantics, but their effectiveness is often constrained by the quality of positive and negative samples. In particular, constructing informative hard negatives in textual data is significantly more difficult than in vision tasks due to the ambiguity and compositionality of natural language. We propose HNCSE, a novel unsupervised contrastive learning framework that enhances sentence representations through hard negative compositional strategies. HNCSE introduces two key components: HNCSE-HNM, which synthesizes informative hard negatives via mixup within the batch, and HNCSE-PM, which generates harder positives by leveraging the most challenging negatives. This joint design improves both alignment and discrimination in embedding space without relying on external supervision. Extensive experiments on semantic textual similarity and transfer tasks demonstrate that HNCSE consistently outperforms existing unsupervised and supervised baselines.}, archive = {J_ASOC}, author = {Wenxiao Liu and Zihong Yang and Chaozhuo Li and Zijin Hong and Jianfeng Ma and Zhiquan Liu and Litian Zhang and Feiran Huang}, doi = {10.1016/j.asoc.2025.113685}, journal = {Applied Soft Computing}, month = {12}, pages = {113685}, shortjournal = {Appl. Soft. Comput.}, title = {Contrastive learning with hard negatives for sentence embeddings}, volume = {184}, year = {2025}, }
• (2025). HumanMoD: A multi-RAG collaborative LLM for inclusive urban public healthcare services. ASOC, 184, 113684. (www)

  @article{ , abstract = {The integration of Large Language Models (LLMs) with healthcare systems offers transformative solutions for sustainable public health programs, especially in underserved urban and rural communities where access to professional medical expertise is limited. This paper introduces HumanMoD, a novel LLM framework designed to emulate collaborative clinical workflows of multi-expert medical doctors. Specifically, the proposed Mixture of Doctors module enables parallel diagnostic streams from diverse medical specialties, mimicking the collaborative decision-making of human doctors to ensure comprehensive assessments in resource-constrained environments. The Knowledge-driven Medical Assistant leverages domain-specific knowledge bases to mitigate LLM hallucinations, ensuring that recommendations are rooted in credible medical knowledge. Exquisitely, the Humanoid Health Conductor and LLM-powered Corrector further refine outputs to minimize diagnostic discrepancies, enhancing the reliability of responses for large-scale public health applications such as community health kiosks and mobile health apps serving remote or low-income areas. Unlike fine-tuned models, HumanMoD operates without parameter adjustment, enabling cost-effective deployment in regions with scarce computational resources, thus bridging the healthcare gap for socially vulnerable groups. Experimental results on MedQA and PubMedQA demonstrate that HumanMoD outperforms state-of-the-art models, highlighting its potential to drive equitable healthcare access and support data-driven public health policies in diverse urban and rural settings.}, archive = {J_ASOC}, author = {Song Sun and Zhijie Zhong and Nanlan Yu and Xinrong Gong and Kaixiang Yang}, doi = {10.1016/j.asoc.2025.113684}, journal = {Applied Soft Computing}, month = {12}, pages = {113684}, shortjournal = {Appl. Soft. Comput.}, title = {HumanMoD: A multi-RAG collaborative LLM for inclusive urban public healthcare services}, volume = {184}, year = {2025}, }
• (2025). A KPI-related fault diagnosis method for multimode manufacturing processes based on supervised minimal gated unit and sparse broad learning system. ASOC, 184, 113679. (www)

  @article{ , abstract = {In multimode manufacturing processes, the key performance indicator (KPI)-related fault diagnosis plays a critical role in ensuring product quality and enhancing economic benefits. However, variability in working modes, lack dynamic description for the KPIs, and model structure redundancy may lead to poor universality and low diagnostic accuracy of conventional deep learning-based approaches. In this work, a supervised minimal gated unit and sparse broad learning system (SMGU-SBLS) is developed for KPI-related fault diagnosis. Specifically, the KPIs and process variables are simultaneously utilized in the SMGU to learn the KPI-related dynamic features. Then, aiming at simplifying the network structure, a sparse version of broad learning system is proposed for fault diagnosis. Furthermore, the expansion capability of SMGU-SBLS has been analyzed. Finally, the proposed SMGU-SBLS network is applied to a real hot strip mill process (HSMP). Simulation results show that the proposed method has higher diagnostic performance than the other four state-of-the-art deep learning methods.}, archive = {J_ASOC}, author = {Chuanfang Zhang and Wenxiao Yin and Chi Zhang and Kaixiang Peng and Xueyi Zhang}, doi = {10.1016/j.asoc.2025.113679}, journal = {Applied Soft Computing}, month = {12}, pages = {113679}, shortjournal = {Appl. Soft. Comput.}, title = {A KPI-related fault diagnosis method for multimode manufacturing processes based on supervised minimal gated unit and sparse broad learning system}, volume = {184}, year = {2025}, }
• (2025). Green maritime transport performance analysis of G-7 countries using an interval-valued fermatean fuzzy ARLON-based decision model. ASOC, 184, 113677. (www)

  @article{ , abstract = {Maritime transport operations are witnessing heightened environmental sensitivities, prompting the implementation of environmentally focused maritime transport practices. The principal aim is to develop a decision support system for evaluating countries' green maritime transport performance. To this end, the interval-valued Fermatean fuzzy (IVFF)-simple weight calculation (SIWEC)-skewness impact through distributional evaluation (SITDE)-alternative ranking using two-step logarithmic normalization (ARLON) model is developed within this study. This method enables the analysis of countries' green maritime transport performance by integrating expert opinions with environmental and maritime transport parameters. The model determines the influence of experts using IVFF sets. It further facilitates the simultaneous use of subjective and objective criteria weighting approaches to compute the weights of green maritime transport performance criteria. ARLON is employed to assess and rank countries' green maritime transport performance levels. The applicability of the IVFF-SIWEC-SITDE-ARLON model is tested through a case study focusing on G-7 countries, and the results supported the successful implementation of the method. Furthermore, sensitivity and comparative analyses demonstrated the consistency and robustness of the model. According to the findings of the case study, the United States emerges as the country with the highest green maritime transport performance among the G-7 countries. The "linear shipping connectivity index" is identified as the most influential criterion in the decision-making process. The study offers actionable recommendations for the maritime industry, thereby contributing to the advancement of green maritime transport practices.}, archive = {J_ASOC}, author = {Galip Cihan Yalçın and Karahan Kara and Emre Kadir Özekenci and Vladimir Simic and Dragan Pamucar}, doi = {10.1016/j.asoc.2025.113677}, journal = {Applied Soft Computing}, month = {12}, pages = {113677}, shortjournal = {Appl. Soft. Comput.}, title = {Green maritime transport performance analysis of G-7 countries using an interval-valued fermatean fuzzy ARLON-based decision model}, volume = {184}, year = {2025}, }
• (2025). Adaptive self-correction network for human motion prediction. ASOC, 184, 113676. (www)

  @article{ , abstract = {Human motion prediction aims to generate future poses from the observed historical human motion sequence. It is fundamental to many intelligent systems, e.g., human–robot interaction and self-driving. Though the existing encoder–decoder methods obtain good performance in some scenarios, there is still a gap between their prediction results and ground truth in many cases. In this work, we propose to estimate the deviation between the decoding results (which will be referred to as the conventional human motion prediction ) and the groundtruth, and integrate this estimation with the conventional prediction results to derive the corrected human motion prediction . In this way, our method can self-correct the conventional prediction results based on a preliminary estimated deviation from it to the groundtruth, and thus enhance the performance. In our work, we adopt five independent lightweight branches rather than a global estimator to estimate the deviation of the five human body components ( i.e., left arm, right arm, torso, left leg, and right leg). Based on this component-wise deviation estimation strategy, we propose a Fixed Self-Correction Network (FSCNet) for human motion prediction to obtain enhanced performance. Recognizing that not all joints exhibit the same motion dynamics inside one given body component, we further propose the Adaptive Self-Correction Network (ASCNet) to let these five estimators adaptively capture the correlated deviations and thus enhance human motion prediction performance. Extensive experiments on three large datasets (Human3.6M, CMU-Mocap, and 3DPW) validate the superiority of our proposed FSCNet and ASCNet over the established works.}, archive = {J_ASOC}, author = {Jinkai Li and Jinghua Wang and Xin Wang and Liang Yan and Xiaoling Luo and Yong Xu}, doi = {10.1016/j.asoc.2025.113676}, journal = {Applied Soft Computing}, month = {12}, pages = {113676}, shortjournal = {Appl. Soft. Comput.}, title = {Adaptive self-correction network for human motion prediction}, volume = {184}, year = {2025}, }
• (2025). RGB-D indoor scene parsing via wavelet sub-band guided transformer. ASOC, 184, 113675. (www)

  @article{ , abstract = {Depth information has been shown to be complementary to RGB scene parsing. However, inherent disparities between RGB and depth modalities create challenges for effective feature fusion. Furthermore, current methods often lose high-frequency information during downsampling, limiting the full utilization of RGB and depth image information. To address these issues, we propose the Wavelet Sub-band Guided Transformer (WSGFormer), which utilizes wavelet sub-band to enhance feature correction, fusion, and refinement. The WSGFormer contains three important modules. Firstly, the Wavelet Cross-attention Rectification Module employs Haar wavelet transforms to decompose features into wavelet sub-band, and adaptively aligns RGB and depth features by extracting their mapping relationships. Secondly, the Multi-scale Fusion Module combines RGB and depth branches, utilizing vertical bar-shaped convolutions to enable cross-modal feature selection and enhance the sensitivity to high-frequency components through frequency-aware techniques. Finally, the Discrepancy Compensation Module starts with high-level semantic information and progressively guides the fusion of adjacent layers downwards, reducing disparities between them through subtraction operations. The evaluation conducted on the NYUv2, SUN-RGBD and ScanNetV2 datasets highlights the superior performance of the proposed WSGFormer.}, archive = {J_ASOC}, author = {Wen Xie and Heng Liu and JiaHao Li}, doi = {10.1016/j.asoc.2025.113675}, journal = {Applied Soft Computing}, month = {12}, pages = {113675}, shortjournal = {Appl. Soft. Comput.}, title = {RGB-D indoor scene parsing via wavelet sub-band guided transformer}, volume = {184}, year = {2025}, }
• (2025). Large language model driven multiple operating conditions identification and predictive control. ASOC, 184, 113674. (www)

  @article{ , abstract = {Dynamic changes in operating conditions are common in industrial process control. Efficient and accurate detection of condition transitions and identification of operating states are critical for achieving precise control across multiple operating conditions. Traditional condition identification methods primarily rely on the numerical similarity of input–output sequences, often neglecting the dynamic semantic information embedded within the data, such as variation rates and overshoot magnitudes, which can lead to frequent misclassifications. Recently, large language models (LLMs) have exhibited remarkable capabilities in semantic understanding of complex sequences, offering a new perspective for the identification of operating conditions. However, their application in industrial control still faces two major challenges. First, the enormous parameter scale of LLMs results in high computational costs, making it difficult to achieve real time comparisons between online sequences and large historical datasets, thereby compromising the timeliness of condition identification. Second, LLMs are designed primarily for textual input, leading to a significant modality gap when processing numerical input–output sequence data, which limits their full semantic understanding potential in industrial scenarios. To address these challenges, this paper proposes a novel large language model driven multiple operating conditions identification and predictive control method. The proposed method fully leverages the semantic understanding capabilities of LLMs by mining the underlying dynamic characteristics of input–output sequences, enabling rapid identification of operating conditions under limited sample scenarios and achieving precise control across multiple conditions. Specifically, first, to accurately detect condition changes in multiple operating condition industrial processes, an Approximate Entropy based operating condition change detection method is proposed. Considering that condition transitions often cause a prediction model mismatch, leading to increased complexity and irregularity in control sequences, approximate entropy is employed to quantify the sequence complexity. A data driven adaptive thresholding mechanism based on kernel density estimation is further developed to achieve robust detection of condition changes. Second, to address the issues of limited samples and low identification accuracy arising from solely relying on numerical features, a novel LLM driven multimodal condition identification method is proposed. This method constructs a sparse representation based prediction model for historical operating conditions, forming a high information density knowledge base consisting of prediction models and representative sequences. To bridge the modality gap between numerical and textual data, a numerical to textual sequence description method enriched with dynamic semantics is innovatively introduced, enabling effective alignment between the two modalities. Furthermore, a dynamic semantics enhanced prompt engineering strategy is developed to fully exploit the LLM’s semantic understanding capabilities, facilitating accurate condition identification even under limited sample conditions. Finally, to achieve rapid response and precise control following condition changes, the proposed method quickly matches a suitable prediction model based on the LLM driven condition identification results and dynamically adjusts control parameters using a rolling horizon optimization strategy, thereby significantly improving control accuracy across multiple operating conditions. Notably, the proposed approach eliminates the need for reconstructing condition models after changes, ensuring smooth and continuous control under dynamic conditions. Extensive experiments verified the superiority of the proposed method in terms of both condition identification and multiple operating conditions control performance.}, archive = {J_ASOC}, author = {Zhongyu Zhang and Minzhi Mao and Keke Huang and Dehao Wu and Chunhua Yang}, doi = {10.1016/j.asoc.2025.113674}, journal = {Applied Soft Computing}, month = {12}, pages = {113674}, shortjournal = {Appl. Soft. Comput.}, title = {Large language model driven multiple operating conditions identification and predictive control}, volume = {184}, year = {2025}, }
• (2025). MetaIndux-PLC: A control logic-guided LLM for PLC code generation in industrial control systems. ASOC, 184, 113673. (www)

  @article{ , abstract = {Programmable Logic Controllers (PLCs) are widely used for automation control, and they are well-suited for industrial systems control tasks. LLMs can assist engineers in streamlining the programming process and reducing development costs, and one of the key issues is the construction of the PLC code dataset. However, the lack of an open-source PLC code dataset in the research community, combined with the high complexity of industrial systems control logic, has caused most LLMs to struggle with generating accurate control code. This complexity arises from the need to manage real-time sensor data fusion, integrate various communication protocols, and ensure compliance with stringent safety and regulatory standards. In this study, we construct ST4Indux, a PLC code dataset specifically for industrial systems control. And we propose the Control Logic-Guided Iterative Fine-Tuning (CLIFT) method, which iteratively optimizes the model’s generation capability. Based on these, we train a large language model named MetaIndux-PLC, to enable the generation of complex motion control code. Additionally, we propose a multi-dimensional evaluation and optimization method to systematically assess the model’s performance in terms of task completion quality, efficiency, and user experience. The experimental results demonstrate that the proposed approach significantly enhances MetaIndux-PLC’s performance and reliability in real-world engineering environments, providing a foundation for the future development of intelligent programming assistance systems.}, archive = {J_ASOC}, author = {Lei Ren and Haotian Wang and Jiabao Dong and Haiteng Wang and Shuai Liu and Yuanjun Laili and Lin Zhang}, doi = {10.1016/j.asoc.2025.113673}, journal = {Applied Soft Computing}, month = {12}, pages = {113673}, shortjournal = {Appl. Soft. Comput.}, title = {MetaIndux-PLC: A control logic-guided LLM for PLC code generation in industrial control systems}, volume = {184}, year = {2025}, }
• (2025). Two-stage identification of false data injection attacks in power systems via semi-supervised deep learning. ASOC, 184, 113672. (www)

  @article{ , abstract = {Previous studies find that malicious data manipulations against the state estimation, such as the false data injection attack (FDIA), can evade the detection of a conventional bad data detector equipped with a measurement meter. This calls for advanced FDIA identification approaches urgently. However, existing data-driven efforts are designed under the assumption that attacks are frequent and the amount of compromised data is comparable to benign data, which may not be realistic. Thus, these approaches deliver unsatisfactory performance under highly imbalanced data in the real world. To overcome this issue, we propose a novel two-stage FDIA identification pipeline, which formulates the problem as global detection and fine-grained localization. Following this framework, we leverage deep support vector data description to distinguish attacks from benign measurements in an unsupervised manner and employ a modified one-dimensional ResNet to locate the attacking aims upon detecting an FDIA. Our approach can overcome existing limitations induced by data-driven methods under infrequent FDIAs, leading to effective and robust FDIA identification. Case studies on IEEE standard 14-bus and 118-bus systems demonstrate the effectiveness and superiority of our approach and validate our findings.}, archive = {J_ASOC}, author = {Fengrui Liu and Keng-Weng Lao and Yida Xu and Yang Li and Haotian Guo and Xiaorui Hu and Yikun Yin}, doi = {10.1016/j.asoc.2025.113672}, journal = {Applied Soft Computing}, month = {12}, pages = {113672}, shortjournal = {Appl. Soft. Comput.}, title = {Two-stage identification of false data injection attacks in power systems via semi-supervised deep learning}, volume = {184}, year = {2025}, }
• (2025). Dynamic target continuous assignment method for unmanned clusters in fragmented information environments. ASOC, 184, 113671. (www)

  @article{ , abstract = {Target assignment stands as a pivotal resource management technique in collaborative clustering. During far-sea search and rescue (SAR) missions, challenges arise in formation control under fragmented information, stemming from unstable communication and incomplete data. Additionally, the dynamic nature of targets poses continuous assignment challenges. Addressing these issues in cluster collaboration represents a significant challenge. This paper proposes a method for dynamic target successive allocation tailored to fragmented information environments. First, inspired by biological cluster behaviors and integrating graph theory and complex network theory, the approach achieves decentered cluster formation by managing subgroup separation and aggregation. This solution effectively addresses the formation control challenge in fragmented information settings during cluster collaboration. Second, leveraging reinforcement learning principles, the method determines device behavioral strategies based on maximizing the device’s interaction rewards with the environment. This approach resolves the dynamic continuous target assignment challenge within clustered environments. This paper proposes a Broken Info-Driven Target Assignment (BI-DTA) method to address these research challenges. Experimental results demonstrate that the method achieves effective decentered formation control and dynamic continuous target assignment rapidly, exhibiting robustness and stability in fragmented information environments.}, archive = {J_ASOC}, author = {Rui Ding and Yuhan Zhu and Baojie Chai}, doi = {10.1016/j.asoc.2025.113671}, journal = {Applied Soft Computing}, month = {12}, pages = {113671}, shortjournal = {Appl. Soft. Comput.}, title = {Dynamic target continuous assignment method for unmanned clusters in fragmented information environments}, volume = {184}, year = {2025}, }
• (2025). A surrogate-assisted evolutionary algorithm based on the two-round selection strategy incorporating local search for expensive high-dimensional multi/many-objective optimization. ASOC, 184, 113670. (www)

  @article{ , abstract = {Surrogate-assisted multi-objective optimization has exhibited excellent performance for solving optimization problems that involve time-intensive computer simulations or resource-intensive physical experiments. However, the majority of existing research has focused on low-dimensional problems. In this paper, a surrogate-assisted evolutionary algorithm based on the two-round selection strategy incorporating local search (TRLS) is proposed for expensive high-dimensional multi/many-objective optimization. Specifically, the convergence and diversity of the trial solutions are assessed based on the estimation of Pareto fronts, and the uncertainty is quantified by analyzing the distribution of relevant points in the decision space. The infill sampling task is guided by the above three performance indicators. Firstly, a preliminary screening is conducted by considering convergence and diversity performance. Subsequently, the quality of candidates is further refined through the implementation of a local search strategy. Finally, a comprehensive fitness is constructed to select the sampling individual. In addition, a dynamic termination criterion is devised for the surrogate-assisted evolution phase. Empirical studies, conducted on two classical benchmark suites and two real-world tasks, reveal the effectiveness and applicability of the proposed TRLS.}, archive = {J_ASOC}, author = {Yang Li and Weigang Li and Songtao Li and Qifeng Wang and Junwei Hu}, doi = {10.1016/j.asoc.2025.113670}, journal = {Applied Soft Computing}, month = {12}, pages = {113670}, shortjournal = {Appl. Soft. Comput.}, title = {A surrogate-assisted evolutionary algorithm based on the two-round selection strategy incorporating local search for expensive high-dimensional multi/many-objective optimization}, volume = {184}, year = {2025}, }
• (2025). A study of the hesitant fuzzy aggregation operators for the transportation of perishable goods under real life scenarios. ASOC, 184, 113668. (www)

  @article{ , abstract = {The transportation of the perishable goods has always been a critical challenge in the market. The inherent susceptibility of the perishable items to spoilage and deterioration during transit necessitates the development and implementation of the preservation technologies that enhance the shelf life and maintain the quality of perishable products during transportation. Also, the carbon emissions stemming from activities associated with transportation present a significant and progressively urgent challenge, markedly contributing to environmental degradation. Consequently, the escalating carbon emissions linked to the transportation-related activities pose a threat to the sustainability of the environment. Recalling these facts, this paper explores the study of an interval valued multi-objective fixed charge solid transportation problem under hesitant fuzzy aggregation operators. Furthermore, this paper introduces a practical mathematical framework designed to represent the decision-making process inherited in the transportation scenarios. The proposed method leverages the hesitant fuzzy aggregation operators to offer a specific approach for the decision-making across such operators. Additionally, it introduces the notion of hesitant degrees for different objectives through the utilization of membership functions. Three conflicting objective functions: time minimization for customer satisfaction, profit maximization for the economic sustainability and minimization of the carbon emissions for the environmental sustainability have been addressed in the suggested model. The transportation time and the fuel consumption have been managed by introducing a variable called the vehicle speed coefficient. An analysis of the deterioration rates of the perishable products has been conducted considering the presence of a preservation value. Moreover, the proposed model has been formulated by employing the diverse approaches and have been solved using multi-objective techniques. A real-life-based numerical problem is presented and solved to validate the proposed concept. The results are compared with respect to different vehicle speeds and preservation values.}, archive = {J_ASOC}, author = {Awdhesh Kumar Bind and Deepika Rani and Ali Ebrahimnejad}, doi = {10.1016/j.asoc.2025.113668}, journal = {Applied Soft Computing}, month = {12}, pages = {113668}, shortjournal = {Appl. Soft. Comput.}, title = {A study of the hesitant fuzzy aggregation operators for the transportation of perishable goods under real life scenarios}, volume = {184}, year = {2025}, }
• (2025). SAGE-net: Single-layer augmented gated encoder network for efficient multimodal sentiment analysis. ASOC, 184, 113665. (www)

  @article{ , abstract = {Multimodal sentiment analysis has achieved remarkable progress. However, the increasing computational complexity of existing models poses significant challenges in resource-constrained scenarios. To address these challenges, this study introduces a single-layer augmented gated encoder network (SAGE-Net), a novel lightweight multimodal sentiment analysis architecture. In contrast to conventional multilayer, deeply stacked architectures, SAGE-Net employs only a single-layer encoder to preserve multimodal feature comprehension, significantly reducing computational complexity. To enable effective inter-modal interaction, a single-layer cross-attention mechanism is integrated. We extensively evaluate multiple feature fusion strategies and data augmentation strategies to enhance model effectiveness. Experimental results on the CMU-MOSI and CMU-MOSEI, and CH-SIMS datasets demonstrate that SAGE-Net achieves competitive performance and significantly lowers model size and tuning costs. These results validate SAGE-Net as a viable lightweight solution for multimodal sentiment analysis in resource-constrained scenarios.}, archive = {J_ASOC}, author = {Jiazheng Zhou and Xin Kang and Weiping Ding and Linhuang Wang and Fei Ding and Kazuyuki Matsumoto and Chenmeng Zhang and Huiwen Chi}, doi = {10.1016/j.asoc.2025.113665}, journal = {Applied Soft Computing}, month = {12}, pages = {113665}, shortjournal = {Appl. Soft. Comput.}, title = {SAGE-net: Single-layer augmented gated encoder network for efficient multimodal sentiment analysis}, volume = {184}, year = {2025}, }
• (2025). Robust weakly supervised product surface defect segmentation based on guided cropping and inpainting extension. ASOC, 184, 113661. (www)

  @article{ , abstract = {In the realm of industrial manufacturing, automatic detection and segmentation of surface defects in products is vital to enhance both product quality and efficiency. However, a large number of existing deep learning methods require a substantial amount of manually labeled data for training, and the high-cost labeling process hampers the practical application of such methods. Towards this end, we present a weakly supervised defect segmentation algorithm without any segment labels. First, a training enhancement method based on a contrastive learning module (CLM) coupled with a guided cropping module (GCM) is proposed to improve the network’s attention to defects in the Defect Focus Classifier (DFC) training phase. Subsequently, a novel inpainting extension module (IEM) generates a final class activation map (CAM) to obtain a pseudo label automatically for segment network training. Finally, conditional random field (CRF) and an additional training round refine the segmentation results. Moreover, the whole process is distilled into a fully supervised segmentation network to improve the inference efficiency. Conducting extensive experimentation, we have achieved 100% and 93.39% average precision (AP) and 41.73% and 53.14% average intersection-over-union (IOU) on the public datasets KolektorSDD and KolektorSDD2, respectively. Furthermore, we verified the generalizability of our method by conducting experiments on several industrial product classes in the MVTec AD, MTD, and DAGM datasets. In these experiments, we achieved favorable classification and segmentation results using solely image classification labels.}, archive = {J_ASOC}, author = {Rui Yan and Xiaojun Wu and Qixun Yang and Michael Yu Wang}, doi = {10.1016/j.asoc.2025.113661}, journal = {Applied Soft Computing}, month = {12}, pages = {113661}, shortjournal = {Appl. Soft. Comput.}, title = {Robust weakly supervised product surface defect segmentation based on guided cropping and inpainting extension}, volume = {184}, year = {2025}, }
• (2025). Semi-supervised lip-tongue segmentation with boundary region contrast sampling. ASOC, 184, 113653. (www)

  @article{ , abstract = {In Traditional Chinese Medicine, the accurate segmentation mask of the tongue and lip is the key of inspection. Although deep learning has made remarkable progress in medical image segmentation, a lot of manual annotations are still required for training. Semi-supervised learning (SSL) is used to reduce annotation work, but its performance often suffers when applied to tongue and lip segmentation, which is because tongue and lip images have noisy background information and unique boundary regions. To alleviate the problem, we propose a semi-supervised framework named Lip-Tongue segmentation with Boundary Region Contrast Sampling (Lip-Tongue-BReCoSample). We first preprocess the data, roughly locating the target and filtering out noisy background information. Then we generate the key boundary regions and sample to carry out contrast learning, which alleviates the problem that SSL cannot make fine modeling of the boundary regions of the target with limited information. After a lot of experiments, our method has achieved good results in SSL, and makes it reach or even exceed the performance of many traditional supervised methods, which can improve MIOU performance from 88.09 to 90.43 (+2.34) in SSL specifically. Our method is also better than the latest large-dataset pre-trained model (e.g., SegGPT). To the best of our knowledge, it is the first application of SSL in tongue and lip semantic segmentation.}, archive = {J_ASOC}, author = {Tao Jiang and Lechao Zhang and Wang Yuan and Liping Tu and Ji Cui and Xiaojuan Hu and Xin Tan and Lizhuang Ma and Jiatuo Xu}, doi = {10.1016/j.asoc.2025.113653}, journal = {Applied Soft Computing}, month = {12}, pages = {113653}, shortjournal = {Appl. Soft. Comput.}, title = {Semi-supervised lip-tongue segmentation with boundary region contrast sampling}, volume = {184}, year = {2025}, }
• (2025). Short-term multistep wind direction prediction of unmanned sailboats based on OVMD and optimized deep learning model. ASOC, 184, 113651. (www)

  @article{ , abstract = {Accurate short-term multistep wind direction prediction is crucial for enhancing the sailing performance and operational safety of unmanned sailboats. Existing methods often face challenges due to the complexity, non-stationarity, and diverse frequency characteristics of wind direction data. In this study, a novel approach is proposed that combines optimal variational mode decomposition (OVMD) with optimized deep learning model for wind direction prediction. First, OVMD is applied to decompose the wind data into stable modal signals, effectively reducing the impact of data complexity and non-stationarity on prediction performance. Considering that different subsequences exhibit distinct frequency patterns, five deep learning models, including Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, are employed to predict each subsequence separately. The most suitable model for each subsequence is selected based on the root mean square error (RMSE) metric. Additionally, hyperparameter optimization is conducted to enhance prediction accuracy, reduce training time, and eliminate the need for subjective parameter settings. Experimental results demonstrate that the proposed method can accurately capture wind direction variations and achieves superior performance compared to baseline models across all evaluation metrics, ensuring high accuracy and stability in short-term multistep wind direction prediction.}, archive = {J_ASOC}, author = {Zhipeng Shen and Shaoqing Zhang and Yang Yang and Zhaoyang Wu and Haomiao Yu}, doi = {10.1016/j.asoc.2025.113651}, journal = {Applied Soft Computing}, month = {12}, pages = {113651}, shortjournal = {Appl. Soft. Comput.}, title = {Short-term multistep wind direction prediction of unmanned sailboats based on OVMD and optimized deep learning model}, volume = {184}, year = {2025}, }
• (2025). Attention-inspired text-based recommender system with explanatory capabilities. ASOC, 184, 113650. (www)

  @article{ , abstract = {Recommender systems are widely used to help users find relevant and personalized items in various domains. However, providing accurate recommendations is not enough to ensure user satisfaction, trust, and engagement. Nowadays, users demand transparency from these systems typically in the form of an explanation of the recommendation given. This paper presents a novel explainable Recommender System designed to generate recommendations from natural language queries while providing model-intrinsic explanations inspired by attention mechanisms. The system adds transparency, interpretability, and new user cold-start capabilities. We evaluate our approach on twelve datasets from diverse domains and languages, demonstrating its effectiveness and robustness. Results show that our proposal achieves competitive accuracy with respect to strong baselines, while consistently outperforming a prior interpretable model developed for the same task.}, archive = {J_ASOC}, author = {Pablo Pérez-Núñez and Paul Buitelaar and Jorge Díez and Oscar Luaces and Antonio Bahamonde}, doi = {10.1016/j.asoc.2025.113650}, journal = {Applied Soft Computing}, month = {12}, pages = {113650}, shortjournal = {Appl. Soft. Comput.}, title = {Attention-inspired text-based recommender system with explanatory capabilities}, volume = {184}, year = {2025}, }
• (2025). MHGCL: A multi-modal hypergraph contrastive learning framework for molecular property prediction. ASOC, 184, 113645. (www)

  @article{ , abstract = {Accurate molecular property prediction (MPP) is pivotal in drug discovery. Although current models integrate multi-modal (1D, 2D, and 3D) molecular features, they often suboptimally leverage pharmacophoric information and inadequately capture higher-order intramolecular relationships, such as conjugated systems. This study introduces a novel multi-modal hypergraph contrastive learning framework (MHGCL) to generate enriched molecular representations for enhanced MPP. MHGCL uniquely employs hypergraphs to model complex, many-to-many interactions within molecules. It incorporates a dual-channel architecture, featuring a hypergraph transformer and an equivariant graph neural network, to distinctly process 2D and 3D molecular information. Crucially, functional group and chemical element-oriented knowledge graphs are integrated to explicitly imbue the model with pharmacophoric knowledge. The contrastive learning strategy effectively aligns these diverse representations. Extensive experiments across ten benchmark datasets demonstrate that MHGCL consistently outperforms existing state-of-the-art methods. Further ablation studies confirm that the proposed hypergraph-based molecular representation captures structural motifs of molecular functional groups more effectively, thereby affirming the design efficacy of the model’s constituent modules.}, archive = {J_ASOC}, author = {Rui Han and Qun Liu and Xu Gong and Guoyin Wang and Li Liu and Xingping Xian}, doi = {10.1016/j.asoc.2025.113645}, journal = {Applied Soft Computing}, month = {12}, pages = {113645}, shortjournal = {Appl. Soft. Comput.}, title = {MHGCL: A multi-modal hypergraph contrastive learning framework for molecular property prediction}, volume = {184}, year = {2025}, }
• (2025). A multi-hop shapley-based framework for graph convolutional network node classification explanation. ASOC, 184, 113615. (www)

  @article{ , abstract = {Graph Convolutional Networks (GCNs) have recently demonstrated superior performance across various graph machine learning tasks. However, they are often regarded as “black-box” models, which has sparked significant interest in developing methods to interpret their predictions. Among various GCN explanation techniques, game-theoretic Shapley value approaches stand out for their ability to identify important nodes, edges, and features. Nonetheless, most Shapley-based explanation models tend to concentrate on dependencies within a fixed radius, resulting in a constrained perceptual field. Additionally, existing explanation methods primarily focus on the total marginal contributions of either very small or very large coalitions during sampling, which limits their effectiveness in capturing the joint marginal contributions inherent in mid-sized subgraphs. To address these challenges, we propose a novel Shapley-based GCN explanation model called MixHopShap, which incorporates multi-hop information and a balanced sampling strategy. MixHopShap employs a multi-hop computational graph construction process to generate an explanation domain, enabling it to capture both local and long-range dependencies. Moreover, MixHopShap introduces a new sampling strategy that promotes balanced coalition coverage, allowing for efficient sampling of mid-sized subgraphs and facilitating learning of marginal contributions for each edge. We conduct experiments on six real world datasets to evaluate the performance of MixHopShap in terms of Fidelity and ROAR metrics. The experimental results demonstrate the superiority of MixHopShap over state-of-the-art methods. Additionally, the explainability of MixHopShap can significantly improve the confidence of GCN’s prediction in node classification tasks.}, archive = {J_ASOC}, author = {Yifan Zheng and Xibei Yang and Qiguo Sun and Keyu Liu and Qihang Guo}, doi = {10.1016/j.asoc.2025.113615}, journal = {Applied Soft Computing}, month = {12}, pages = {113615}, shortjournal = {Appl. Soft. Comput.}, title = {A multi-hop shapley-based framework for graph convolutional network node classification explanation}, volume = {184}, year = {2025}, }
• (2025). Graph neural network-based interactive clustering enhanced by human knowledge. ASOC, 184, 113595. (www)

  @article{ , abstract = {Clustering techniques face persistent challenges in balancing automation with human interpretability. Traditional methods require laborious parameter tuning and domain expertise to define similarity measures and validate results, while deep learning approaches trade transparency for performance. To bridge this gap, we propose a human-in-the-loop framework that synergizes domain knowledge with graph-based semi-supervised learning. Our system enables users to iteratively refine clusters through intuitive visual adjustments on a subset of data, guided by real-time quality metrics to reduce errors and decision fatigue. These sparse annotations propagate to unlabeled instances via a graph neural network (GNN) that models latent relationships through modularity-driven structural learning. By translating cluster adjustments into semi-supervised classification tasks, our method eliminates manual feature engineering and scales to large datasets without retraining. Evaluations on two subsets of the MNIST dataset demonstrated that the NMI (Normalized Mutual Information) of our method improved by 50.44% and 64.77% relative to baseline clustering method, respectively.}, archive = {J_ASOC}, author = {Yunzhe Wang and Yushi Li and Qiming Fu and Chengtao Ji and You Lu and Jianping Chen}, doi = {10.1016/j.asoc.2025.113595}, journal = {Applied Soft Computing}, month = {12}, pages = {113595}, shortjournal = {Appl. Soft. Comput.}, title = {Graph neural network-based interactive clustering enhanced by human knowledge}, volume = {184}, year = {2025}, }
• (2025). Advanced deep learning approaches for fault detection in solar PV systems: A comparative study of SPDA and AIFD-SolDL. ASOC, 184, 113592. (www)

  @article{ , abstract = {Efficient operation of solar photovoltaic (PV) systems is critical for maximizing power generation and ensuring optimal energy conversion. However, faults in PV modules can significantly impact system performance and reduce energy output. Therefore, accurate identification and diagnosis of these malfunctions are essential. To address the challenge of fault detection in solar PV systems, this study presents two distinct approaches. The first, called Solar Panel Degradation Assessment (SPDA), evaluates faults in solar panels by analyzing degradation effects while considering environmental factors like radiation and temperature. The second approach, named Efficient Ensemble Deep Learning Model for Enhancing Fault Detection in Solar PV Systems (AIFD-SolDL), utilizes advanced deep learning techniques, including DenseNet201, Inception-ResNet-v2, and Inception-v3, to enhance fault detection accuracy. In the AIFD-SolDL approach, PV module data undergo deep feature extraction followed by dimensionality reduction using principal component analysis (PCA). The reduced feature set is then used to train classifiers such as support vector machines (SVM), Gaussian Naive Bayes (GaussianNB), and random forests (RF) to differentiate between normal and fault conditions. Performance metrics, including precision, accuracy, recall, and F1-score, are computed for each combination of feature extractor and classifier. Extensive experiments with both the Solar Panel Images Dataset and the Infrared Solar Module Dataset show that the proposed approaches outperform state-of-the-art methods. For instance, the AIFD-SolDL approach utilizing SVM achieved perfect accuracy, precision, recall, and F1-score of 100% on the Solar Panel Images Dataset. Overall, the SPDA approach effectively detects faults, while deep learning techniques demonstrate high accuracy in fault classification, thereby enhancing the reliability of PV system maintenance and optimization.}, archive = {J_ASOC}, author = {Mohamed R. Shoaib and Heba M. Emara and Jun Zhao and Milad Taleby Ahvanooey and Essam Nabil}, doi = {10.1016/j.asoc.2025.113592}, journal = {Applied Soft Computing}, month = {12}, pages = {113592}, shortjournal = {Appl. Soft. Comput.}, title = {Advanced deep learning approaches for fault detection in solar PV systems: A comparative study of SPDA and AIFD-SolDL}, volume = {184}, year = {2025}, }
• (2025). Neuroevolution-based multiobjective algorithm for feature selection and binary classification of DNA microarrays. ASOC, 184, 113520. (www)

  @article{ , abstract = {The genomes of organisms have been sequenced for many years, leading to the discovery of thousands of genes. DNA microarrays are widely used tools for simultaneously analysing numerous genes, commonly employed in detecting and identifying various diseases, including cancer. However, microarray datasets have non-relevant and redundant information, hindering their analysis. This problem is further exacerbated considering the datasets’ high dimensionality and imbalanced classes. Consequently, standard practice involves incorporating a feature selection process to identify the most relevant genes and their associations with diseases. Various methods have been employed to address this task. However, none have taken a more holistic approach that effectively handles feature selection, automatically identifies the optimal classifier configuration, and manages potential conflicting objectives simultaneously. In response, this study introduces the S -metric selection - multiobjective neuroevolution of augmenting topologies (SMS-MONEAT) algorithm, which combines the multiobjective optimisation framework from S -metric selection - evolutionary multiobjective algorithm (SMS-EMOA) and the evolutionary operators from the neuroevolution algorithm N3O, a variation from NEAT which stands for ‘3 new operators’. SMS-MONEAT algorithm was designed to perform both feature selection and optimise the configuration of artificial neural networks for classification tasks. SMS-MONEAT was evaluated against classic and state-of-the-art methods for feature selection and microarray classification. The experiments were conducted on 20 highly challenging cancer-type datasets primarily sourced from the Curated Microarray Database, and the results were investigated for statistical significance. The findings suggest that SMS-MONEAT either outperforms or achieves competitive results in terms of classification compared to the mentioned methods, while at the same time, it selects a smaller subset of features.}, archive = {J_ASOC}, author = {Daniel García-Núñez and Katya Rodrígez-Vázquez and Carlos Hernández and Edgar Galván}, doi = {10.1016/j.asoc.2025.113520}, journal = {Applied Soft Computing}, month = {12}, pages = {113520}, shortjournal = {Appl. Soft. Comput.}, title = {Neuroevolution-based multiobjective algorithm for feature selection and binary classification of DNA microarrays}, volume = {184}, year = {2025}, }
• (2025). Deep learning-based bangla text normalization with emotion classification for expressive text-to-speech synthesizer. ASOC, 184, 112899. (www)

  @article{ , abstract = {This paper presents a novel method for text normalization and emotion classification specifically designed for the preprocessing steps of Expressive Text-to-Speech (ETTS) synthesizers. Text normalization, which converts non-standard words into standardized forms, is essential for generating clear and coherent output in ETTS synthesizers. Previous studies have encountered difficulties with Bangla homographic words, prompting the development of our proposed method. Our approach begins with the creation of a tokenized and categorized dataset consisting of 26 unique semiotic classes, utilizing a rule-based method with regular expressions. This dataset is derived from the Bangla Text Normalization Corpus (BTN Corpus), which contains 2 million Bangla sentences. Initially, Bangla written texts are tokenized, and each token is classified using a Temporal Convolutional Network (TCN) algorithm trained on the BTN Corpus to identify its semiotic class. This classification aids in generating normalized text, where the resulting tokens are reassembled into coherent sentences for final output. In addition to text normalization, we implement emotion classification for each normalized sentence using a Hierarchical Attention Network (HAN) model. The HAN model was trained on 67,277 normalized texts from the Bangla Normalized Emotion Text Corpus (BNET Corpus), categorizing each text into one of six emotion classes through a rule-based method with regular expressions. The proposed method demonstrates high accuracy rates, achieving a token classification accuracy of 99.977 % with the TCN and an emotion classification accuracy of 99.735 % with the HAN model.}, archive = {J_ASOC}, author = {Md. Rezaul Islam and Mohammad Shahidur Rahman}, doi = {10.1016/j.asoc.2025.112899}, journal = {Applied Soft Computing}, month = {12}, pages = {112899}, shortjournal = {Appl. Soft. Comput.}, title = {Deep learning-based bangla text normalization with emotion classification for expressive text-to-speech synthesizer}, volume = {184}, year = {2025}, }

AUTOM - 58

• (2026). Congealed neural network design for uncertain nonlinear spatiotemporal control systems. AUTOM, 183, 112636. (www)

  @article{ , abstract = {This note is concerned with the approximation-based adaptive control problem for a class of uncertain nonlinear spatiotemporal systems. A congealed neural network (ConNN) is first proposed to approximate nonlinear spatiotemporal uncertainties arising from system states and time-varying parameters. Unlike conventional NN approximation structures, the ConNN explicitly decomposes the time-varying coupling weight into a congealed weight and a time-dependent perturbation. The congealed weight is estimated using a standard adaptive law for constant parameters, while the residual perturbation is handled within the network structure. To enhance robustness, smooth sliding-mode-like functions are then embedded into the control architecture, effectively attenuating bias terms, particularly in reference tracking scenarios. It is shown that the resulting ConNN-based adaptive controller guarantees adjustable, bounded-error tracking performance, thereby extending the applicability of robust adaptive control to complex spatiotemporal systems and outperforming existing NN-based approaches.}, archive = {J_AUTOM}, author = {Tianrun Liu and Yang-Yang Chen and Xiaohua Ge}, doi = {10.1016/j.automatica.2025.112636}, journal = {Automatica}, month = {1}, pages = {112636}, shortjournal = {Automatica}, title = {Congealed neural network design for uncertain nonlinear spatiotemporal control systems}, volume = {183}, year = {2026}, }
• (2026). Distributed adaptive fixed-time formation tracking for heterogeneous multi-agent systems. AUTOM, 183, 112632. (www)

  @article{ , abstract = {This article investigates the problem of fixed-time time-varying formation tracking control (FTC) for heterogeneous linear multiagent systems (MASs) under the directed communication graph. It is assumed that the Laplacian matrix associated with the communication graph is unavailable and that the system matrices of the leader are only available to its neighboring followers. This differs from many existing works on fixed-time FTC problems where the communication graphs are typically undirected and protocol designs often rely on certain global information. A novel distributed observer is first put forward to estimate both the state and system matrices of the leader in fixed time. Then, an adaptive scheme is developed to solve the time-varying regulator equations resulting from the estimated leader system matrices in fixed time. Based on the proposed observer and the adaptive solutions to the regulator equations, a distributed adaptive fixed-time FTC protocol is further proposed via coordinate transformation techniques. It is shown that our proposed controllers do not require the input matrices of the followers to be of full row rank. It is also shown that the concerned fixed-time FTC problem can be solved with the proposed fixed-time FTC strategy in a distributed manner. Our results can be directly applied to solve both the adaptive fixed-time cooperative output regulation problem and the leader-following consensus problems of MASs under the directed graph. Finally, the effectiveness of the proposed fixed-time FTC strategy is demonstrated through a numerical example.}, archive = {J_AUTOM}, author = {Shiyu Zhou and Dong Sun and Gang Feng}, doi = {10.1016/j.automatica.2025.112632}, journal = {Automatica}, month = {1}, pages = {112632}, shortjournal = {Automatica}, title = {Distributed adaptive fixed-time formation tracking for heterogeneous multi-agent systems}, volume = {183}, year = {2026}, }
• (2026). Nature-inspired dynamic control for pursuit-evasion of robots. AUTOM, 183, 112629. (www)

  @article{ , abstract = {The pursuit-evasion problem is widespread in nature, engineering, and societal applications. It is commonly observed in nature that predators often exhibit faster speeds than their prey but have less agile maneuverability. Over millions of years of evolution, animals have developed effective and efficient strategies for pursuit and evasion. In this paper, we provide a dynamic framework for the pursuit-evasion problem of unicycle systems, drawing inspiration from nature. First, we address the scenario with one pursuer and one evader by proposing an Alert-Turn control strategy, which consists of two efficient ingredients: a sudden turning maneuver and an alert condition for starting and maintaining the maneuver. We present and analyze the escape and capture results at two levels: a lower level of a single run and a higher level with respect to parameters’ changes. In addition, we provide a theorem with sufficient conditions for capture. The Alert-Turn strategy is then extended to more complex scenarios involving multiple pursuers and evaders by integrating aggregation control laws and a target-changing mechanism. By adjusting a ‘selfish parameter’, the aggregation control commands produce various escape patterns of evaders: cooperative mode, selfish mode, and their combinations. The influence of the selfish parameter is quantified, and the target-changing mechanism is explored from a statistical perspective. Our findings align closely with observations in nature. Finally, the proposed strategies are validated through numerical simulations that replicate some chasing behaviors of animals in nature.}, archive = {J_AUTOM}, author = {Panpan Zhou and Sirui Li and Benyun Zhao and Bo Wahlberg and Xiaoming Hu}, doi = {10.1016/j.automatica.2025.112629}, journal = {Automatica}, month = {1}, pages = {112629}, shortjournal = {Automatica}, title = {Nature-inspired dynamic control for pursuit-evasion of robots}, volume = {183}, year = {2026}, }
• (2026). Event/self-triggered cooperative control via reinforcement learning for a quadrotor team under multiple faults and denial-of-service attacks. AUTOM, 183, 112628. (www)

  @article{ , abstract = {The cooperative fault-tolerant control problem of a quadrotor team, subject to multiple faults and denial-of-service (DoS) attacks, is addressed via event/self-triggered strategies and reinforcement learning. Multiple under-actuated quadrotors with nonlinearities, couplings, and unknown dynamical parameters are introduced to achieve distributed cooperation. Event-triggered observers are developed to estimate position references under uncertain cyber faults and DoS attacks. Observer-based optimal controllers for the nominal position and attitude subsystems are learned by off-policy reinforcement learning without dynamical information. Fault-tolerant controllers are constructed by integrating the learned controllers and adaptive actuator fault compensators. A self-triggered strategy is further proposed to avoid continuous communication. The relationship among the time-varying topology coupled with cyber faults, the features of DoS attacks, and the closed-loop control system stability is analyzed, and the exclusion of Zeno behavior is proved. Simulation results illustrate the effectiveness of the proposed methods.}, archive = {J_AUTOM}, author = {Ziming Ren and Hao Liu and Hongwei Zhang and Ci Chen and Frank L. Lewis}, doi = {10.1016/j.automatica.2025.112628}, journal = {Automatica}, month = {1}, pages = {112628}, shortjournal = {Automatica}, title = {Event/self-triggered cooperative control via reinforcement learning for a quadrotor team under multiple faults and denial-of-service attacks}, volume = {183}, year = {2026}, }
• (2026). Adaptive distributed observer design for nonlinear multiagent systems. AUTOM, 183, 112625. (www)

  @article{ , abstract = {Distributed state estimation is crucial for the leader-following control problem of multiagent systems (MASs). In this paper, adaptive distributed observers (DOs) are designed for a nonlinear autonomous leader system using only its output. Via system transformation, the DO design problem of the origin system is converted to that of a canonical system with lumped dynamics. When the lumped dynamics is parametric, an adaptive DO is developed to reconstruct the state and the unknown parameters under an undirected topology, which also addresses the distributed state/parameter estimation problem of an uncertain autonomous system with its dynamics in a parametric form. Then, the DO framework is extended to the case of non-parametric uncertainties, and a neural network (NN) DO is designed for reconstructing the state/lumped dynamics over a strongly connected digraph Finally, the effectiveness of the proposed DOs is demonstrated via numerical simulations.}, archive = {J_AUTOM}, author = {Jixing Lv and Changhong Wang and Lihua Xie}, doi = {10.1016/j.automatica.2025.112625}, journal = {Automatica}, month = {1}, pages = {112625}, shortjournal = {Automatica}, title = {Adaptive distributed observer design for nonlinear multiagent systems}, volume = {183}, year = {2026}, }
• (2026). A novel switching rule to observer-based control for switched systems. AUTOM, 183, 112622. (www)

  @article{ , abstract = {This study explores observer-based control for switched systems subject to partially unknown sojourn probability information. To more accurately capture the dynamic nature of switched systems, a novel switching rule related to duration time is devised, in which the system switching is governed by a joint distribution function. This function uniquely integrates both the current system mode and its duration, effectively addressing the challenges of incompleteness in probability information. Distinct from the conventional semi-Markov kernel technique, the sojourn probabilities involve fewer parameters and can be more readily measured using statistical techniques, thereby lessening computational burden. Additionally, a non-monotonic Lyapunov function is constructed, leading to less conservative conditions that ensure the mean-square stability of the switched systems. Eventually, the effectiveness and superiority of the devised methodology are verified through a practical example.}, archive = {J_AUTOM}, author = {Jun Cheng and Tianfeng Tang and Huaicheng Yan and Zheng-Guang Wu and Dan Zhang}, doi = {10.1016/j.automatica.2025.112622}, journal = {Automatica}, month = {1}, pages = {112622}, shortjournal = {Automatica}, title = {A novel switching rule to observer-based control for switched systems}, volume = {183}, year = {2026}, }
• (2026). Sampled-data primal–dual gradient dynamics in model predictive control. AUTOM, 183, 112621. (www)

  @article{ , abstract = {Model predictive control (MPC) can incur computational burden that exceeds what the application warrants, sometimes even for linear systems. Recently, a rapid computation method that guides the input toward convergence with the optimal control problem solution by employing primal–dual gradient (PDG) dynamics has been proposed for linear MPCs. However, stability has been ensured under the assumption that the controller is a continuous-time system, leading to potential instability when the controller undergoes discretization and is implemented as a sampled-data system. In this paper, we propose a discrete-time dynamical controller incorporating specific modifications to the PDG approach and present stability conditions. Additionally, we introduce an extension to enhance control performance that was traded off in the original. Numerical examples substantiate that our proposed method, which can be executed in only 1 μ s on a commodity laptop, not only ensures stability considering sampled-data implementation but also substantially enhances the control performance.}, archive = {J_AUTOM}, author = {Ryuta Moriyasu and Sho Kawaguchi and Kenji Kashima}, doi = {10.1016/j.automatica.2025.112621}, journal = {Automatica}, month = {1}, pages = {112621}, shortjournal = {Automatica}, title = {Sampled-data primal–dual gradient dynamics in model predictive control}, volume = {183}, year = {2026}, }
• (2026). Cooperative dynamic surrounding coverage control for multi-autonomous surface vehicle fleets with limited sensing ranges. AUTOM, 183, 112620. (www)

  @article{ , abstract = {Cooperative surrounding coverage by a multi-autonomous surface vehicle (ASV) fleet involves dynamic coverage with concentric moving trajectories. This low-cost form of coverage, which operates with limited perceptual ranges, has long presented a significant challenge in real-world collective marine detection applications. To this end, the paper establishes a surrounding coverage control framework that minimizes the required number of ASVs by covering a designated region across multiple consecutive periods. A distributed randomized gradient-free (DRGF) protocol is thereby proposed to identify the optimal central point of the coverage region. Accordingly, a distributed control law is designed to regulate the ASV dynamic surrounding coverage operation around the optimal concentric point. Significantly, sufficient conditions are derived to guarantee the asymptotical convergence of the closed-loop multi-ASV fleet governed by the present DGRF. Finally, experiments are conducted on a self-established lake-based multi-ASV fleet platform to verify the effectiveness of the present surrounding coverage scheme.}, archive = {J_AUTOM}, author = {Jiayu Zou and Hai-Tao Zhang and Bin Liu and Xiaohua Liu and Jianing Ding and Ning Xing and Shiqi Fu and Lijun Zhu}, doi = {10.1016/j.automatica.2025.112620}, journal = {Automatica}, month = {1}, pages = {112620}, shortjournal = {Automatica}, title = {Cooperative dynamic surrounding coverage control for multi-autonomous surface vehicle fleets with limited sensing ranges}, volume = {183}, year = {2026}, }
• (2026). Asynchronous resilient collision-free formation control for nonlinear MASs under DoS attacks. AUTOM, 183, 112619. (www)

  @article{ , abstract = {The multi-agent systems (MASs) exchange the communication information by the communication network, which is vulnerable to denial-of-service (DoS) attacks. DoS attacks can disrupt the communication channels and result in the collisions between agents, thus it is an important and challenging topic to study the collision avoidance formation control issue under DoS attacks. This paper investigates the asynchronous resilient collision-free formation control issue for nonlinear MASs with DoS attacks. An asynchronous resilient distributed observer is proposed to estimate the output information of leader under DoS attacks. Based on the designed distributed observer and by constructing two high-order filters, an asynchronous resilient collision-free formation control method is presented by backstepping control design theory. The presented formation control method can ensure that the formation tracking errors asymptotically converge to zero, and the collision avoidance objective is realized. Moreover, the non-differentiable problem of virtual controllers is solved. Finally, we apply the developed formation control method to multiple Euler-Lagrangian (EL) systems, the simulation and comparison results verify its effectiveness.}, archive = {J_AUTOM}, author = {Jun Zhang and Jun Ning and Shaocheng Tong}, doi = {10.1016/j.automatica.2025.112619}, journal = {Automatica}, month = {1}, pages = {112619}, shortjournal = {Automatica}, title = {Asynchronous resilient collision-free formation control for nonlinear MASs under DoS attacks}, volume = {183}, year = {2026}, }
• (2026). Sector stabilization criterion of a novel nonlinear flexible marine riser coupled system. AUTOM, 183, 112618. (www)

  @article{ , abstract = {This paper establishes a sector stabilization criterion for a nonlinear flexible marine riser system that incorporates lateral and transverse coupling vibrations, derived from Hamilton’s principle. This criterion, grounded in the sector-bounded condition, encompasses a wide range of linear and nonlinear feedback control laws applied to the transverse and lateral directions at the top boundary of the flexible marine riser, respectively. In the analysis, the nonlinear semigroup theory is utilized to establish the well-posedness of the resulting closed-loop coupled system. Notably, the solution demonstrates continuous dependence on the initial conditions. Furthermore, the exponential stability of the closed-loop coupled system is achieved by employing a generalized Gronwall-type integral inequality and the integral multiplier method, which involves the innovative development of an energy-like functional. To demonstrate the effectiveness of the proposed controls, numerical simulations utilizing the finite element method are presented.}, archive = {J_AUTOM}, author = {Yi Cheng and Xin Wang and Yuhu Wu and Bao-Zhu Guo}, doi = {10.1016/j.automatica.2025.112618}, journal = {Automatica}, month = {1}, pages = {112618}, shortjournal = {Automatica}, title = {Sector stabilization criterion of a novel nonlinear flexible marine riser coupled system}, volume = {183}, year = {2026}, }
• (2026). Global exponential stabilization of 2 × 2 linear hyperbolic PDEs via dynamic event-triggered backstepping control. AUTOM, 183, 112617. (www)

  @article{ , abstract = {This paper introduces novel dynamic event-triggered control (ETC) mechanisms for 2 × 2 linear hyperbolic PDEs in three configurations: continuous-time event-triggered control (CETC), periodic event-triggered control (PETC), and self-triggered control (STC). These mechanisms ensure global exponential stability (GES) under ETC using PDE backstepping, with stability estimates provided in the spatial L 2 norm of the states. The proposed CETC and PETC designs are observer-based and require continuous boundary measurements collocated with the actuation. In contrast, the STC design requires full-state measurements; however, unlike CETC and PETC, it does not require continuous measurements for the triggering mechanism—only measurements taken at event times. In the CETC design, a lower bound on the time between two consecutive events is enforced, and a dynamic variable with appropriately designed switching dynamics is introduced. By employing a novel Lyapunov functional, GES of the closed-loop system is established under zero-order hold implementation of the backstepping control between events. Events are triggered when the dynamic variable crosses zero from the positive side, after which it is immediately reset to an appropriate nonnegative value. Detecting events, therefore, necessitates continuous monitoring of this dynamic variable. To address this limitation, PETC and STC strategies are proposed. The PETC design identifies a suitable triggering condition that requires only periodic checks and derives an upper bound on the allowable sampling period. This PETC approach preserves the GES guaranteed by CETC without requiring continuous monitoring of a triggering condition, although it still relies on continuous measurements. Unlike CETC and PETC, STC requires neither continuous measurements nor monitoring of a triggering condition. Instead, at each event, STC computes the time to the next event — beyond a suitably enforced minimal dwell-time — using only measurements taken at events. Despite relying solely on event-triggered measurements, STC is capable of guaranteeing GES of the closed-loop system. The well-posedness of the closed-loop systems under all three strategies is established. A simulation study is provided to illustrate the theoretical results.}, archive = {J_AUTOM}, author = {Bhathiya Rathnayake and Mamadou Diagne}, doi = {10.1016/j.automatica.2025.112617}, journal = {Automatica}, month = {1}, pages = {112617}, shortjournal = {Automatica}, title = {Global exponential stabilization of 2 × 2 linear hyperbolic PDEs via dynamic event-triggered backstepping control}, volume = {183}, year = {2026}, }
• (2026). Signal temporal logic control synthesis among uncontrollable dynamic agents with conformal prediction. AUTOM, 183, 112616. (www)

  @article{ , abstract = {The control of dynamical systems under temporal logic specifications among uncontrollable dynamic agents is challenging due to the agents’ a-priori unknown behavior. Existing works have considered the problem where either all agents are controllable, the agent models are deterministic and known, or no safety guarantees are provided. We propose a predictive control synthesis framework that guarantees, with high probability, the satisfaction of signal temporal logic (STL) tasks that are defined over a controllable system in the presence of uncontrollable stochastic agents. We use trajectory predictors and conformal prediction to construct probabilistic prediction regions for each uncontrollable agent that are valid over multiple future time steps. Specifically, we construct a normalized prediction region over all agents and time steps to reduce conservatism and increase data efficiency. We then formulate a worst-case bilevel mixed integer program (MIP) that accounts for all agent realizations within the prediction region to obtain an open-loop controller that provably guarantee task satisfaction with high probability. To efficiently solve this bilevel MIP, we propose an equivalent MIP program based on KKT conditions of the original bilevel formulation. Building upon this, we design a closed-loop controller, where both recursive feasibility and task satisfaction can be guaranteed with high probability. We illustrate our control synthesis framework on two case studies.}, archive = {J_AUTOM}, author = {Xinyi Yu and Yiqi Zhao and Xiang Yin and Lars Lindemann}, doi = {10.1016/j.automatica.2025.112616}, journal = {Automatica}, month = {1}, pages = {112616}, shortjournal = {Automatica}, title = {Signal temporal logic control synthesis among uncontrollable dynamic agents with conformal prediction}, volume = {183}, year = {2026}, }
• (2026). Event-triggered boundary estimation for 2 × 2 hyperbolic PDEs with disturbance. AUTOM, 183, 112615. (www)

  @article{ , abstract = {This paper addresses the event-triggered boundary estimation problem for 2 × 2 linear hyperbolic partial differential equations (PDEs) subject to disturbances. These disturbances enter the PDEs through the boundary and are represented as an ordinary differential equation (ODE). The scenario considered involves transmitting boundary measurements to the observer only when necessary, as determined by a dynamic event-triggering condition. This approach aims to conserve communication and computational resources. The event-triggered observer is designed using the backstepping method, and an event-triggering condition is proposed to determine the time instants when sampled measurements should be transmitted. Under this event-triggered estimation mechanism, a minimal dwell-time between consecutive triggering time instants is guaranteed. Furthermore, the well-posedness of the solutions and the exponential convergence of the estimation error are ensured. Finally, the obtained results are applied to the Saint-Venant equations to demonstrate the effectiveness of the event-triggered estimation method.}, archive = {J_AUTOM}, author = {Yan Zhao and Rafael Vazquez}, doi = {10.1016/j.automatica.2025.112615}, journal = {Automatica}, month = {1}, pages = {112615}, shortjournal = {Automatica}, title = {Event-triggered boundary estimation for 2 × 2 hyperbolic PDEs with disturbance}, volume = {183}, year = {2026}, }
• (2026). Distributed data-driven unknown-input observers. AUTOM, 183, 112614. (www)

  @article{ , abstract = {Unknown inputs related to, e.g., sensor aging, modeling errors, or device bias, represent a major concern in wireless sensor networks, as they degrade the state estimation performance. To improve the performance, unknown-input observers (UIOs) have been proposed. Most of the results available to design UIOs are based on explicit system models, which can be difficult or impossible to obtain in real-world applications. Data-driven techniques, on the other hand, have become a viable alternative for the design and analysis of unknown systems using only data. In this context, a novel data-driven distributed unknown-input observer (D-DUIO) for unknown continuous-time linear time-invariant (LTI) systems is developed, which requires solely some data collected offline, without any prior knowledge of the system matrices. In the paper, first, a model-based approach to the design of a DUIO is presented. A sufficient condition for the existence of such a DUIO is recalled, and a new one is proposed, that is prone to a data-driven adaptation. Moving to a data-driven approach, it is shown that under suitable assumptions on the input/output/state data collected from the continuous-time system, it is possible to both claim the existence of a D-DUIO and to derive its matrices in terms of the matrices of pre-collected data. Finally, the efficacy of the D-DUIO is illustrated by means of numerical examples.}, archive = {J_AUTOM}, author = {Yuzhou Wei and Giorgia Disarò and Wenjie Liu and Jian Sun and Maria Elena Valcher and Gang Wang}, doi = {10.1016/j.automatica.2025.112614}, journal = {Automatica}, month = {1}, pages = {112614}, shortjournal = {Automatica}, title = {Distributed data-driven unknown-input observers}, volume = {183}, year = {2026}, }
• (2026). Robustness of supervisory controllers subject to measurement disturbances. AUTOM, 183, 112613. (www)

  @article{ , abstract = {This paper studies the supervisory control problem for a class of uncertain nonlinearly parameterized systems in the presence of measurement disturbance. Based on the well-established estimator-based supervisory control structure, a robustification of supervisory controller dealing with measurement disturbance is developed taking advantage of the monitoring signals redesign, such that the plant state asymptotically converges to a given set-point or its neighborhood, subject to the vanishing or the persistent measurement disturbance, respectively. Moreover, a constructive design of the multi-estimator using measurement feedback is proposed for the supervisory control of a class of uncertain nonlinearly parameterized systems in the strict-feedback form. A numerical simulation based on an uncertain mass–spring system is given to show the efficacy of our proposed algorithm, in which we use an event-trigger to be a measurement disturbance generator.}, archive = {J_AUTOM}, author = {Yutian Wang and Qingkai Meng and Yi Jiang}, doi = {10.1016/j.automatica.2025.112613}, journal = {Automatica}, month = {1}, pages = {112613}, shortjournal = {Automatica}, title = {Robustness of supervisory controllers subject to measurement disturbances}, volume = {183}, year = {2026}, }
• (2026). Data-driven min–max MPC for linear systems: Robustness and adaptation. AUTOM, 183, 112612. (www)

  @article{ , abstract = {Data-driven controllers design is an important research problem, in particular when data is corrupted by the noise. In this paper, we propose a data-driven min–max model predictive control (MPC) scheme using noisy input-state data for unknown linear time-invariant (LTI) system. The unknown system matrices are characterized by a set-membership representation using the noisy input-state data. Leveraging this representation, we derive an upper bound on the worst-case cost and determine the corresponding optimal state-feedback control law through a semidefinite program (SDP). We prove that the resulting closed-loop system is robustly stabilized and satisfies the input and state constraints. Further, we propose an adaptive data-driven min–max MPC scheme which exploits additional online input-state data to improve closed-loop performance. Numerical examples show the effectiveness of the proposed methods.}, archive = {J_AUTOM}, author = {Yifan Xie and Julian Berberich and Frank Allgöwer}, doi = {10.1016/j.automatica.2025.112612}, journal = {Automatica}, month = {1}, pages = {112612}, shortjournal = {Automatica}, title = {Data-driven min–max MPC for linear systems: Robustness and adaptation}, volume = {183}, year = {2026}, }
• (2026). Local practically safe extremum seeking with assignable rate of attractivity to the safe set. AUTOM, 183, 112611. (www)

  @article{ , abstract = {We present Assignably Safe Extremum Seeking (ASfES), an algorithm designed to minimize a measured, static objective function while maintaining a measured, static metric of safety (a control barrier function or CBF) to be positive in a practical sense. We ensure that for trajectories with safe initial conditions, the violation of safety can be made arbitrarily small through appropriately chosen design constants. We also guarantee an assignable “attractivity” rate: from unsafe initial conditions, the trajectories approach the safe set, in the sense of the measured CBF, at a rate no slower than a user-assigned rate. Similarly, from safe initial conditions, the trajectories approach the unsafe set, in the sense of the CBF, no faster than the assigned attractivity rate. The feature of assignable attractivity is not present in the semiglobal version of safe extremum seeking, where the semiglobality of convergence is achieved by slowing the adaptation. We also demonstrate local convergence of the parameter to a neighborhood of the minimum of a quadratic objective function constrained to the safe set with a linear CBF. The ASfES algorithm and analysis are multivariable, but we also extend the algorithm to a Newton-Based ASfES scheme which we show is only useful in the scalar case. The proven properties of the designs are illustrated through simulation examples.}, archive = {J_AUTOM}, author = {Alan Williams and Miroslav Krstic and Alexander Scheinker}, doi = {10.1016/j.automatica.2025.112611}, journal = {Automatica}, month = {1}, pages = {112611}, shortjournal = {Automatica}, title = {Local practically safe extremum seeking with assignable rate of attractivity to the safe set}, volume = {183}, year = {2026}, }
• (2026). Secondary safety control for systems with sector bounded nonlinearities. AUTOM, 183, 112610. (www)

  @article{ , abstract = {We consider the problem of safety verification and safety-aware controller synthesis for systems with sector bounded nonlinearities. We aim to keep the states of the system within a given safe set under potential actuator and sensor attacks. Specifically, we adopt the setup that a controller has already been designed to stabilize the plant. Using invariant sets and barrier certificate theory, we first give sufficient conditions to verify the safety of the closed-loop system under attacks. Furthermore, by using a subset of sensors that are assumed to be free of attacks, we provide a synthesis method for a secondary controller that enhances the safety of the system. The sufficient conditions to verify safety are derived using Lyapunov-based tools and the S -procedure. Using the projection lemma, the conditions are then formulated as linear matrix inequality (LMI) problems which can be solved efficiently. Lastly, our theoretical results are illustrated through numerical simulations.}, archive = {J_AUTOM}, author = {Yankai Lin and Michelle S. Chong and Carlos Murguia}, doi = {10.1016/j.automatica.2025.112610}, journal = {Automatica}, month = {1}, pages = {112610}, shortjournal = {Automatica}, title = {Secondary safety control for systems with sector bounded nonlinearities}, volume = {183}, year = {2026}, }
• (2026). Non-parametric IQC multipliers in data-driven robust controller synthesis. AUTOM, 183, 112608. (www)

  @article{ , abstract = {The paper presents a robust data-driven controller synthesis method for generalised multi-input multi-output (MIMO) systems. Using the frequency response of a linear time-invariant (LTI) MIMO system and characterising perturbations through Integral Quadratic Constraint (IQC), the method provides a convex set of controllers robust to perturbations. This facilitates the design of controllers with either robust stability or robust performance criteria. Notably, the proposed method is versatile, as it is also applicable for non-parametric IQC multipliers. An example of a non-parametric IQC multiplier for elliptical uncertainty quantification is demonstrated and subsequently employed in designing a robust controller for a hybrid active-passive micro-vibration platform. Experimental results show that the synthesised controller effectively achieves the desired levels of both robustness and performance.}, archive = {J_AUTOM}, author = {Vaibhav Gupta and Elias Klauser and Alireza Karimi}, doi = {10.1016/j.automatica.2025.112608}, journal = {Automatica}, month = {1}, pages = {112608}, shortjournal = {Automatica}, title = {Non-parametric IQC multipliers in data-driven robust controller synthesis}, volume = {183}, year = {2026}, }
• (2026). Iterative learning control for performance-driven switched systems under all unknown channel gains. AUTOM, 183, 112607. (www)

  @article{ , abstract = {Fading channels introduce random and unknown gains into signal transmission, which has a great impact on the accuracy of the transmitted data, thus affecting system stability. However, there are few studies on this regard in the control field. In this paper, the transmission of both feedback and switching signals is affected by the unknown channel gains, which induces severe mode asynchronization in the switched systems under iterative learning control (ILC). For the faded switching signals, by designing a calibration-rounding mechanism (CRM) to recover it, the asynchronous time can gradually decrease with the iteration, and finally the system achieves synchronization, that is calibration-rounding-based iterative synchronization (CRIS). For the faded feedback signals, an improved gain estimation mechanism (IGEM) is given to correct the signal faster with fewer iterations. Moreover, a novel tracking performance-driven switching law (TPD-SL) is proposed to reasonably schedule subsystems for obtaining optimal performance. A set of new average dwell time (ADT) conditions with the minimum synchronization time is obtained. Finally, the effectiveness of the method is verified by numerical simulations.}, archive = {J_AUTOM}, author = {Yiwen Qi and Ziyu Qu and Dong Shen}, doi = {10.1016/j.automatica.2025.112607}, journal = {Automatica}, month = {1}, pages = {112607}, shortjournal = {Automatica}, title = {Iterative learning control for performance-driven switched systems under all unknown channel gains}, volume = {183}, year = {2026}, }
• (2026). Induced norm analysis of linear systems for nonnegative input signals. AUTOM, 183, 112606. (www)

  @article{ , abstract = {This paper is concerned with the analysis of the L p ( p ∈ [ 1 , ∞ ) , p = ∞ ) induced norms of continuous-time linear systems where input signals are restricted to be nonnegative. This norm is referred to as the L p + induced norm in this paper. It has been shown recently that the L 2 + induced norm is effective for the stability analysis of nonlinear feedback systems where the nonlinearity returns only nonnegative signals. However, the exact computation of the L 2 + induced norm is essentially difficult. To get around this difficulty, in the first part of this paper, we provide a copositive-programming-based method for the upper bound computation by capturing the nonnegativity of the input signals by copositive multipliers. In the second part, we consider how far the L 2 + induced norm can be smaller than the standard L 2 induced norm, and derive the uniform infimum on the ratio of the L 2 + induced norm to the L 2 induced norm over all linear systems including infinite-dimensional ones. Then, for each linear system, we finally derive a computation method of the lower bounds of the L 2 + induced norm that are larger than (or equal to) the value determined by the uniform infimum. The effectiveness of the upper/lower bound computation methods is illustrated by numerical examples.}, archive = {J_AUTOM}, author = {Yoshio Ebihara and Noboru Sebe and Hayato Waki and Dimitri Peaucelle and Sophie Tarbouriech and Victor Magron and Tomomichi Hagiwara}, doi = {10.1016/j.automatica.2025.112606}, journal = {Automatica}, month = {1}, pages = {112606}, shortjournal = {Automatica}, title = {Induced norm analysis of linear systems for nonnegative input signals}, volume = {183}, year = {2026}, }
• (2026). Distributed fault-tolerant control for a class of uncertain cascaded ODE-PDE multi-agent systems. AUTOM, 183, 112605. (www)

  @article{ , abstract = {In this paper, a distributed fault-tolerant control is proposed for a class of uncertain cascaded ordinary differential equation (ODE)-partial differential equation (PDE) multi-agents. For the cascaded system under consideration, the output of ODEs serves as the input signal of PDEs such that a distributed control involving states of the ODE as well as the PDE system is expected to be designed for consensus tracking. Based on the designed distributed control, where the actuator is assumed to be healthy, a distributed fault-tolerant control is designed to guarantee the performance of the system, and it is further improved to deal with actuator failures. To enhance the robustness of the cascaded system, the system uncertainties and the external disturbances of both followers and leader are compensated using adaptive laws and an adaptive distributed observer, based on which the stabilization of the cascaded ODE-PDE multi-agent system can be achieved with partially known agents’ information. Simulation results are carried out to prove the effectiveness of the proposed method.}, archive = {J_AUTOM}, author = {Xueyan Xing and Guoqiang Hu}, doi = {10.1016/j.automatica.2025.112605}, journal = {Automatica}, month = {1}, pages = {112605}, shortjournal = {Automatica}, title = {Distributed fault-tolerant control for a class of uncertain cascaded ODE-PDE multi-agent systems}, volume = {183}, year = {2026}, }
• (2026). A compensation-oriented algorithm for difference-driven identification under binary-valued observations and data packet dropout. AUTOM, 183, 112604. (www)

  @article{ , abstract = {This paper investigates the identification problem for finite impulse response (FIR) systems with binary-valued observations under event-triggered communication mechanism and data packet dropout. The challenge lies in the inability to distinguish between untriggered events and packet loss when no information is received, which prevents us from obtaining the statistical properties of the binary-valued sequence. A compensation-oriented difference-driven identification (CODD) algorithm is proposed to estimate the parameter by recovering the mean of the original binary-valued sequence, where different values for the observation estimates are assigned when receiving 0, 1 or nothing. Even though, the convergence analysis of the parameter estimate is still challenging since the assigned values are dependent. To tackle this difficulty, the estimate error is divided into two parts: an initial assigned value related part, which is demonstrated to be convergent through the construction of an auxiliary set, and the remaining component, which happens to be a convergent martingale-difference sequence. As a result, the almost sure convergence and the asymptotic normality of the CODD algorithm are established when data packet loss probability is less than 1 2 . By calculating the communication rate, it is proven that the difference-driven mechanism can save 50% of the communication cost compared to original binary-valued systems. Furthermore, when data packet loss probability is high, an m -channel compensation-oriented identification ( m -CODD) algorithm is constructed by utilizing retransmission of the each observation for m times, which is designed based on the packet loss probability. The properties of m -CODD algorithm including convergence, asymptotic normality and communication rate are established. Numerical simulations are illustrated to show the theoretical results.}, archive = {J_AUTOM}, author = {Tianning Han and Ying Wang and Jin Guo and Yanlong Zhao}, doi = {10.1016/j.automatica.2025.112604}, journal = {Automatica}, month = {1}, pages = {112604}, shortjournal = {Automatica}, title = {A compensation-oriented algorithm for difference-driven identification under binary-valued observations and data packet dropout}, volume = {183}, year = {2026}, }
• (2026). Adaptive fault-tolerant control of nonlinear systems: A self-regulating spatiotemporal performance approach. AUTOM, 183, 112602. (www)

  @article{ , abstract = {This paper investigates the problem of performance constraints of uncertain nonlinear systems with actuator faults. Particularly, by developing the boundary threshold triggering strategy, the performance boundaries would be updated adaptively once the distance between the tracking error and the performance boundaries is smaller than the given threshold. Moreover, compared with the existing works, the transient behaviors are improved, and the limitation imposed on the initial conditions is removed attributed to the construction of novel performance functions and error transformation. Then, an adaptive fault-tolerant control scheme with self-regulating spatiotemporal performance is designed for a class of nonlinear systems with non-parametric uncertainties. It is shown that both the boundedness of the closed-loop signals and the satisfactory performance constraints are guaranteed in the presence of unpredictable actuator failure. The effectiveness of the proposed method is verified by theoretical analysis and numerical simulation.}, archive = {J_AUTOM}, author = {Zeqiang Li and Jason J.R. Liu and Yongduan Song}, doi = {10.1016/j.automatica.2025.112602}, journal = {Automatica}, month = {1}, pages = {112602}, shortjournal = {Automatica}, title = {Adaptive fault-tolerant control of nonlinear systems: A self-regulating spatiotemporal performance approach}, volume = {183}, year = {2026}, }
• (2026). Fully coupled nonlinear FBS△Es: Solvability and LQ control insights. AUTOM, 183, 112601. (www)

  @article{ , abstract = {In this paper, a class of fully coupled nonlinear forward–backward stochastic difference equations (FBS △ Es) is proposed and the existence of solutions is proved based on a linear-quadratic (LQ) optimal control problem. Inspired from the solvability studies of various forward–backward stochastic differential equations (FBSDEs), the dominant-monotone framework is discretized and a continuum approach is used to prove the unique solvability of the fully coupled FBS △ Es and to obtain a pair of estimates on the solutions, and finally, the conclusions are applied to the related LQ problem.}, archive = {J_AUTOM}, author = {Zhipeng Niu and Qingxin Meng and Xun Li and Maoning Tang}, doi = {10.1016/j.automatica.2025.112601}, journal = {Automatica}, month = {1}, pages = {112601}, shortjournal = {Automatica}, title = {Fully coupled nonlinear FBS△Es: Solvability and LQ control insights}, volume = {183}, year = {2026}, }
• (2026). Aerial target encirclement and interception with noisy range observations. AUTOM, 183, 112600. (www)

  @article{ , abstract = {This paper proposes a strategy to encircle and intercept a non-cooperative aerial point-mass moving target by leveraging noisy range measurements for state estimation. In this approach, the guardians actively ensure the observability of the target by using an anti-synchronization (AS), 3D “vibrating string” trajectory, which enables rapid position and velocity estimation based on the Kalman filter. Additionally, a novel anti-target controller is designed for the guardians to enable adaptive transitions from encircling a protected target to encircling, intercepting, and neutralizing a hostile target, taking into consideration the input constraints of the guardians. Based on the guaranteed uniform observability, the exponentially bounded stability of the state estimation error and the convergence of the encirclement error are rigorously analyzed. Simulation results and real-world UAV experiments are presented to further validate the effectiveness of the system design.}, archive = {J_AUTOM}, author = {Fen Liu and Shenghai Yuan and Thien-Minh Nguyen and Wei Meng and Lihua Xie}, doi = {10.1016/j.automatica.2025.112600}, journal = {Automatica}, month = {1}, pages = {112600}, shortjournal = {Automatica}, title = {Aerial target encirclement and interception with noisy range observations}, volume = {183}, year = {2026}, }
• (2026). Stochastic gradient hamiltonian sequential monte carlo filter with earth mover’s distance sampling for target tracking. AUTOM, 183, 112599. (www)

  @article{ , abstract = {High-speed missile tracking requires advanced filtering algorithms to overcome the limitations of traditional nonlinear methods. This study presents the stochastic gradient Hamiltonian sequential Monte Carlo filter, combining stochastic gradient Hamilton Monte Carlo with sequential Monte Carlo (SMC) for enhanced sampling performance and reduced computational burden. The method incorporates Earth Mover’s Distance based adaptive resampling with theoretical bounds for optimal particle count. Validation through univariate nonstationary growth model simulation and bearing-only tracking experiments demonstrates superior performance over conventional methods, achieving 15 % root mean square error improvement compared to conventional SMC and 30 % over extended Kalman filter/unscented Kalman filter approaches.}, archive = {J_AUTOM}, author = {Chang Ho Kang and Sun Young Kim}, doi = {10.1016/j.automatica.2025.112599}, journal = {Automatica}, month = {1}, pages = {112599}, shortjournal = {Automatica}, title = {Stochastic gradient hamiltonian sequential monte carlo filter with earth mover’s distance sampling for target tracking}, volume = {183}, year = {2026}, }
• (2026). Distributed nash equilibrium seeking with a dynamic set of players. AUTOM, 183, 112598. (www)

  @article{ , abstract = {This paper formulates a new distributed Nash equilibrium seeking problem with a dynamic set of players in which players are allowed to join and leave the network in a free manner during the decision-making process. To accommodate the dynamic joining and leaving behaviors of the players, a status estimation mechanism, which is capable of estimating in a finite time whether the players are active or inactive, is introduced. Based on the status estimation mechanism, a gradient play based algorithm is developed for distributed Nash equilibrium seeking in the dynamic environment. It is shown that under strongly connected communication graphs, players’ actions are convergent to a small neighborhood of the new Nash equilibrium linearly every time the player set changes. Moreover, the convergence accuracy and convergence rate can be adjusted by suitably tuning the step-size. To cover more general communication scenarios, strongly connected graphs are further relaxed to be B-jointly connected graphs, under which the convergence properties of the proposed algorithm are analytically studied. Furthermore, the upper bound of the average tracking error is quantified to evaluate the dynamic performance of the proposed algorithm. In the last, a simulation study on energy consumption games is given to verify the effectiveness of the proposed algorithm.}, archive = {J_AUTOM}, author = {Yuxuan Liu and Maojiao Ye and Lei Ding and Lihua Xie and Qing-Long Han}, doi = {10.1016/j.automatica.2025.112598}, journal = {Automatica}, month = {1}, pages = {112598}, shortjournal = {Automatica}, title = {Distributed nash equilibrium seeking with a dynamic set of players}, volume = {183}, year = {2026}, }
• (2026). Constrained finite-time and fixed-time stabilization for linear systems: Adaptive implicit lyapunov function-based control. AUTOM, 183, 112597. (www)

  @article{ , abstract = {In this paper, finite-time and fixed-time stabilization problems are investigated for single-input single-output (SISO) linear system under the control input constraint. The achievement of finite-time or fixed-time convergence rates is facilitated through the utilization of adaptive implicit Lyapunov function (ILF)-based control. For ease of practical implementation, the dynamics of Approximated-ILF (AILF) guarantees the precise estimation of ILF, while the stability of AILF-based control holds established. Furthermore, from both performance and input-constrained safety considerations, the anti-windup (AW) AILF endows the system with tolerance to saturation and maintains the non-asymptotic convergence properties. Numerical simulations support the obtained theoretical results and verify their effectiveness.}, archive = {J_AUTOM}, author = {Peng Wang and Mou Chen and Shuzhi Sam Ge and Xiaobing Zhang}, doi = {10.1016/j.automatica.2025.112597}, journal = {Automatica}, month = {1}, pages = {112597}, shortjournal = {Automatica}, title = {Constrained finite-time and fixed-time stabilization for linear systems: Adaptive implicit lyapunov function-based control}, volume = {183}, year = {2026}, }
• (2026). Input-to-state stability of self-triggered impulsive control systems. AUTOM, 183, 112596. (www)

  @article{ , abstract = {This paper studies the local input-to-state stability ( LISS ) of nonlinear systems under the self-triggered impulsive control ( STIC ) method. A novel LISS -type comparison principle is proposed, by which the LISS property of addressed system can be derived by the LISS property of its comparison system. On the basis of it, some Lyapunov-based sufficient conditions for non-Zeno behavior and LISS of nonlinear systems are provided in the framework of STIC . Moreover, the designed self-triggering mechanism ( STM ) is in the form of an explicit relationship with simple structure and east implementation. Finally, two numerical examples are given to illustrate the effectiveness of the proposed results.}, archive = {J_AUTOM}, author = {Xiaodi Li and Mingzhu Wang}, doi = {10.1016/j.automatica.2025.112596}, journal = {Automatica}, month = {1}, pages = {112596}, shortjournal = {Automatica}, title = {Input-to-state stability of self-triggered impulsive control systems}, volume = {183}, year = {2026}, }
• (2026). Event-triggered global and group consensus for linear multiagent systems with multiplicative noises. AUTOM, 183, 112595. (www)

  @article{ , abstract = {This article is concerned with the event-triggered global and group consensus problems for linear multiagent systems (MASs) with multiplicative noises and directed topology. First, for the global consensus of MASs under a balanced directed graph, a stochastic input-to-state stability property is proved to show that the piecewise constant control input has sufficient feedback capability; based on the relative state measurement, two novel static and dynamic event-triggered mechanisms (ETMs) are established to reduce the number of controller updates, where the Zeno phenomenon is effectively eliminated by imposing a uniformly strictly fixed positive time for all the inter-event times. Especially, new stochastic stability analysis techniques are developed to give sufficient conditions of mean square and almost sure consensus in both static and dynamic ETMs. Moreover, the obtained results on global consensus are extended to the setting of group consensus in order to cope with some grouping tasks in practical applications. Finally, some examples are presented to show the validity of theoretical results.}, archive = {J_AUTOM}, author = {Ruru Jia and Yuan-Hua Ni and Guangchen Wang}, doi = {10.1016/j.automatica.2025.112595}, journal = {Automatica}, month = {1}, pages = {112595}, shortjournal = {Automatica}, title = {Event-triggered global and group consensus for linear multiagent systems with multiplicative noises}, volume = {183}, year = {2026}, }
• (2026). Data-model-hybrid-driven near-optimal operational control of two-time-scale industrial systems with unknown operational model. AUTOM, 183, 112594. (www)

  @article{ , abstract = {This article discusses the optimal operational control (OOC) problem of two-time-scale (TTS) industrial systems with unknown operation model. Based on the singular perturbation theory (SPT), the OOC problem of TTS industrial systems is decomposed into an optimal regulation problem in fast time-scale and an optimal set-point tracking problem in slow time-scale. Then, by convex duality, the obtained optimization problems are equivalently transformed into convex optimization (CO) problems, and a data-model-hybrid-driven composite controller is designed. The design method of this composite controller avoids the potential numerical stiffness problems, and does not need complete system dynamics information while ensuring the steady-state output tracking error converges to zero. Finally, an example of mixed separation thickening process (MSTP) of hematite beneficiation is given to show the effectiveness of the proposed scheme.}, archive = {J_AUTOM}, author = {Yao Xu and Linna Zhou and Jianguo Zhao and Lei Ma and Chunyu Yang}, doi = {10.1016/j.automatica.2025.112594}, journal = {Automatica}, month = {1}, pages = {112594}, shortjournal = {Automatica}, title = {Data-model-hybrid-driven near-optimal operational control of two-time-scale industrial systems with unknown operational model}, volume = {183}, year = {2026}, }
• (2026). Event-triggered distributed optimal coordination of heterogeneous linear MASs over weight-unbalanced digraphs. AUTOM, 183, 112593. (www)

  @article{ , abstract = {In this paper, the distributed optimal coordination problem of heterogeneous linear multi-agent systems is studied, in which the actual information of the left eigenvector and out-degree cannot be obtained in advance. The underlying communication network among agents is assumed to be a weight-unbalanced and strongly connected digraph. The global objective function of the whole system is the sum of privately-known local objective functions, which are assumed to be strongly convex with Lipschitz continuous gradients. To achieve the distributed optimal coordination with low communication and computational resource consumption, and to efficiently reduce the frequency of control signal updates, a new distributed event-triggered control scheme is proposed. In this scheme, an auxiliary system with adaptive edge weights is established to generate the optimal trajectories and the tracking control inputs are designed for agents to track the given optimal trajectories. It is privacy-preserving to some extent as no actual state or output information of agents is communicated with neighbors. Sufficient conditions are derived, under which the distributed optimal coordination under consideration can be exponentially achieved. Moreover, the Zeno-behavior is shown to be excluded. Additionally, for cases where agent states are unmeasurable, a new distributed event-triggered observer-based control input is designed for each agent to achieve optimal coordination. Finally, numerical simulations are carried out to show the good performance of the proposed control scheme.}, archive = {J_AUTOM}, author = {Dandan Wang and Jialing Zhou and Guanghui Wen}, doi = {10.1016/j.automatica.2025.112593}, journal = {Automatica}, month = {1}, pages = {112593}, shortjournal = {Automatica}, title = {Event-triggered distributed optimal coordination of heterogeneous linear MASs over weight-unbalanced digraphs}, volume = {183}, year = {2026}, }
• (2026). Reducing real-time complexity via sub-control lyapunov functions: From theory to experiments. AUTOM, 183, 112592. (www)

  @article{ , abstract = {The techniques to design control Lyapunov functions (CLF), along with a proper stabilizing feedback, possibly in the presence of constraints, often provide control laws that are too complex for proper implementation online, especially when an optimization problem is involved. In this work, we show how to acquire an alternative, computationally attractive feedback. Given a nominal CLF and a nominal state feedback, we say that a different positive definite function is a Sub-control Lyapunov function (SCLF) if its Lyapunov derivative is negative-definite and bounded above by the Lyapunov derivative of the nominal function with the nominal control. It turns out that if we consider a family of basis functions, then an SCLF can be computed by linear programming, with an infinite number of constraints. The idea is that although the offline computational burden to achieve the new controller and solve the linear program is considerable, the online computational burden is drastically reduced. Comprehensive simulations and experiments on drone control are conducted to demonstrate the effectiveness of the study.}, archive = {J_AUTOM}, author = {Huu-Thinh Do and Franco Blanchini and Stefano Miani and Ionela Prodan}, doi = {10.1016/j.automatica.2025.112592}, journal = {Automatica}, month = {1}, pages = {112592}, shortjournal = {Automatica}, title = {Reducing real-time complexity via sub-control lyapunov functions: From theory to experiments}, volume = {183}, year = {2026}, }
• (2026). Adversarial dynamic games for markov jump systems: A policy iteration Q-learning method. AUTOM, 183, 112591. (www)

  @article{ , abstract = {This paper presents a reinforcement Q-learning approach for solving adversarial dynamic games in Markov jump systems. The H ∞ control problem is first formulated as a two-player zero-sum dynamic game, where the control policy and the disturbance policy act as adversarial players. To derive the Nash equilibrium control strategies for such games, a set of coupled algebraic Riccati equations is established, with the disturbance attenuation level properly prescribed. On this basis, two novel data-driven parallel Q-learning algorithms are proposed. The advantages of the proposed method are threefold: (i) it does not require precise knowledge of the system dynamics; (ii) it learns the optimal disturbance attenuation level; (iii) it yields Nash equilibrium control strategies. Finally, two simulation examples validate the effectiveness of the proposed method.}, archive = {J_AUTOM}, author = {Hao Shen and Jiacheng Wu and Jing Wang and Zhengguang Wu}, doi = {10.1016/j.automatica.2025.112591}, journal = {Automatica}, month = {1}, pages = {112591}, shortjournal = {Automatica}, title = {Adversarial dynamic games for markov jump systems: A policy iteration Q-learning method}, volume = {183}, year = {2026}, }
• (2026). H2/H∞ state and output feedback control with sparse actuation. AUTOM, 183, 112581. (www)

  @article{ , abstract = {In this paper, we present novel convex optimization formulations for designing full-state and output-feedback controllers with sparse actuation that achieve user-specified H 2 and H ∞ performance criteria. The sparsity is induced through the ℓ 1 -minimization over channel-wise H 2 norms from disturbances to the individual actuator signals, while simultaneously constraining H 2 or H ∞ norm from disturbances to the output variables The proposed approach is applied to a structural dynamics problem, demonstrating the advantages of simultaneous optimization of the control law and the actuation architecture in realizing an efficient closed-loop system, as well as highlighting the trade-offs between maximum allowable actuator magnitudes and the controller sparsity.}, archive = {J_AUTOM}, author = {Vedang M. Deshpande and Raktim Bhattacharya}, doi = {10.1016/j.automatica.2025.112581}, journal = {Automatica}, month = {1}, pages = {112581}, shortjournal = {Automatica}, title = {H2/H∞ state and output feedback control with sparse actuation}, volume = {183}, year = {2026}, }
• (2026). Adaptive dynamic event–triggered distributed optimal coordination of heterogeneous uncertain nonlinear multiagent systems. AUTOM, 183, 112580. (www)

  @article{ , abstract = {This paper addresses the distributed optimal coordination problem for a class of heterogeneous uncertain nonlinear multiagent systems. Instead of relying on the analytical forms of gradient functions, we use the measured gradient values depending on agents’ real-time outputs and propose a novel adaptive distributed control scheme. This scheme integrates event-triggered optimal coordinators, high-order filters, and tracking controllers. To handle the interaction between optimal coordinators and filters, we incorporate a new compensation term into the updating law for the coupling weight of each edge. Moreover, we design a novel adaptive distributed dynamic event-triggering mechanism that ensures that the inter-event times of each agent are lower bounded by a positive constant. Asymptotic convergence of agents’ outputs to the optimal point is proved by constructing a composite Lyapunov function. The proposed control scheme does not depend on global topology information. A numerical example is given to demonstrate the effectiveness of the proposed control scheme.}, archive = {J_AUTOM}, author = {Tianyu Liu and Lu Liu}, doi = {10.1016/j.automatica.2025.112580}, journal = {Automatica}, month = {1}, pages = {112580}, shortjournal = {Automatica}, title = {Adaptive dynamic event–triggered distributed optimal coordination of heterogeneous uncertain nonlinear multiagent systems}, volume = {183}, year = {2026}, }
• (2026). Accelerated primal–dual methods for strongly convex objective functions in continuous and discrete time. AUTOM, 183, 112579. (www)

  @article{ , abstract = {In this paper, we introduce a “second-order primal” + “first-order dual” continuous-time dynamic for linearly constrained optimization problems, where the objective function is μ -strongly convex. We consider a constant damping 2 μ for the second-order ordinary differential equation in the primal variable, following Nesterov’s acceleration for strongly convex optimization. A positive constant scaling is applied to the primal variable, while a positive increasing scaling function is applied to the dual variable. We prove that the proposed dynamic achieves a fast convergence rate for both the objective residual and the feasibility violation, with the decay rate potentially reaching O ( e − μ t ) . Additionally, we show that the dynamic is robust under small perturbations. By discretizing the proposed continuous-time dynamic, we develop an accelerated linearized augmented Lagrangian method for strongly convex composite optimization with linear constraints, where the objective function has a nonsmooth + smooth composite structure. The proposed algorithm achieves a fast convergence rate that matches the one of the continuous-time dynamic. We also consider an inexact version of the proposed algorithm, which can be viewed as a discrete version of the perturbed continuous-time dynamic. Numerical results are provided to verify the practical performances.}, archive = {J_AUTOM}, author = {Xin He and Dong He and Ya-Ping Fang}, doi = {10.1016/j.automatica.2025.112579}, journal = {Automatica}, month = {1}, pages = {112579}, shortjournal = {Automatica}, title = {Accelerated primal–dual methods for strongly convex objective functions in continuous and discrete time}, volume = {183}, year = {2026}, }
• (2026). Hierarchical supervisory control of networked and cyber-attacked discrete-event systems. AUTOM, 183, 112578. (www)

  @article{ , abstract = {In standard supervisory control of discrete-event systems, partial (incomplete) observations are given by deterministic functions such as natural projections, which erase unobservable events, or masks, which can represent indistinguishable events, where two or more different events yield the same observation. However, communication channels in modern technological systems are not always reliable and can be attacked by malicious external agents. In that case, the plant observations obtained by the supervisor may not be deterministic, e.g., due to delays and losses, or external attacks. This paper considers a unified supervisory control framework with set-valued (nondeterministic) observations and proposes a simplified version of nondeterministic observability, together with a generalized normality. It shows how the results of hierarchical control can be extended to the networked and cyber-attacked discrete-event systems at the same time.}, archive = {J_AUTOM}, author = {Shaowen Miao and Jan Komenda and Feng Lin}, doi = {10.1016/j.automatica.2025.112578}, journal = {Automatica}, month = {1}, pages = {112578}, shortjournal = {Automatica}, title = {Hierarchical supervisory control of networked and cyber-attacked discrete-event systems}, volume = {183}, year = {2026}, }
• (2026). On the design of linear time varying model predictive control for trajectory stabilization. AUTOM, 183, 112577. (www)

  @article{ , abstract = {Stabilizing a reference trajectory of a nonlinear system is a recurrent, non-trivial task in control engineering. A common approach is to linearize the dynamics along the trajectory, thus deriving a linear-time-varying (LTV) model, and to design a model predictive controller (MPC), which results to be computationally efficient, since only convex programs need to be solved in real time, while retaining constraint handling capabilities. Building on recent developments in gain-scheduling control design, where linearization errors and tracking error bounds are considered, a new approach to derive such LTV-MPC controllers is presented. The method addresses the systematic derivation of a suitable terminal cost. The resulting MPC law is tube-based, exploiting the co-designed auxiliary gain-scheduled controller. Computational and implementation aspects are discussed as well, and the resulting hierarchical method is demonstrated both in simulation and in experiments with a small drone with fast dynamics and limited embedded computational capacity.}, archive = {J_AUTOM}, author = {Nicolas Kessler and Lorenzo Fagiano}, doi = {10.1016/j.automatica.2025.112577}, journal = {Automatica}, month = {1}, pages = {112577}, shortjournal = {Automatica}, title = {On the design of linear time varying model predictive control for trajectory stabilization}, volume = {183}, year = {2026}, }
• (2026). Intersection-based architectures for decentralized diagnosis of discrete event systems. AUTOM, 183, 112576. (www)

  @article{ , abstract = {In this paper, two intersection-based architectures, named the normal-state-estimator-intersection-based architecture (N-SEI architecture) and the failure-state-estimator-intersection-based architecture (F-SEI architecture), are examined for decentralized diagnosis of discrete event systems. For each of these architectures, the corresponding notion of codiagnosability is defined. These defined notions of codiagnosability are incomparable with inference diagnosability for the inference-based architecture. In addition, codiagnosability for the N-SEI architecture is weaker than the existing notion of intersection-based codiagnosability, while codiagnosability for the F-SEI architecture is incomparable with it. For each of the N-SEI and F-SEI architectures, a method for verifying the corresponding notion of codiagnosability is developed.}, archive = {J_AUTOM}, author = {Shigemasa Takai and Takashi Yamamoto}, doi = {10.1016/j.automatica.2025.112576}, journal = {Automatica}, month = {1}, pages = {112576}, shortjournal = {Automatica}, title = {Intersection-based architectures for decentralized diagnosis of discrete event systems}, volume = {183}, year = {2026}, }
• (2026). Distributed stochastic constrained optimization with constant step-sizes via saddle-point dynamics. AUTOM, 183, 112575. (www)

  @article{ , abstract = {This paper considers distributed stochastic optimization problems over a multi-agent network, where each agent collaboratively minimizes the sum of individual expectation-valued cost functions subject to nonidentical set constraints. We first recast the distributed constrained optimization as a constrained saddle-point problem. Subsequently, two distributed stochastic algorithms via optimistic gradient descent ascent (SOGDA) and extragradient (SEG) methods are developed with constant step sizes, in which the variable sample-size technique is incorporated to reduce the variance of the sampled gradients. We present the explicit selection criteria of the constant step size, under which the developed algorithms achieve almost sure convergence to an optimal solution. Moreover, the convergence rate is O ( 1 / k ) for merely convex cost functions, which matches the optimal rate of its deterministic counterpart. Finally, a numerical example is provided to reflect the theoretical findings.}, archive = {J_AUTOM}, author = {Yi Huang and Shisheng Cui and Xianlin Zeng and Ziyang Meng}, doi = {10.1016/j.automatica.2025.112575}, journal = {Automatica}, month = {1}, pages = {112575}, shortjournal = {Automatica}, title = {Distributed stochastic constrained optimization with constant step-sizes via saddle-point dynamics}, volume = {183}, year = {2026}, }
• (2026). A characterization method of terminal ingredients for nonlinear MPC using value-based reinforcement learning. AUTOM, 183, 112574. (www)

  @article{ , abstract = {The stability of nonlinear model predictive control (MPC) relies significantly on stabilizing factors such as the terminal region and cost. A larger terminal region not only expands the region of attraction for the closed-loop system but also contributes to reducing online computation costs. However, existing methods in the literature often impose limitations on the degrees of freedom available for characterizing terminal ingredients. This limitation arises from the reliance on either a predetermined linear local controller or a preset control Lyapunov function. This paper introduces an innovative approach to terminal ingredient characterization leveraging value-based reinforcement learning (RL). This method provides ample degrees of freedom for expanding the terminal region. To achieve this, a deep neural network is employed to learn the parametric state value function, serving as the terminal cost for MPC. The local controller adopts a one-step MPC instead of a predetermined linear or nonlinear feedback controller. Subsequently, a terminal set sequence is constructed iteratively through the one-step set expansion. The proposed approach’s effectiveness is validated through simulations.}, archive = {J_AUTOM}, author = {Jinghan Cui and Jinwu Gao and Xiangjie Liu and Yuqi Liu and Shuyou Yu}, doi = {10.1016/j.automatica.2025.112574}, journal = {Automatica}, month = {1}, pages = {112574}, shortjournal = {Automatica}, title = {A characterization method of terminal ingredients for nonlinear MPC using value-based reinforcement learning}, volume = {183}, year = {2026}, }
• (2026). Aperiodic-sampled neural network controllers with closed-loop stability verifications. AUTOM, 183, 112573. (www)

  @article{ , abstract = {In this paper, we synthesize two aperiodic-sampled deep neural network (DNN) control schemes, based on the closed-loop tracking stability guarantees. By means of the integral quadratic constraint coping with the input–output behavior of system uncertainties/nonlinearities and the convex relaxations of nonlinear DNN activations leveraging their local sector-bounded attributes, we establish conditions to design the event- and self-triggered logics and to compute the ellipsoidal inner approximations of region of attraction, respectively. Finally, we perform a numerical example of an inverted pendulum to illustrate the effectiveness of the proposed aperiodic-sampled DNN control schemes.}, archive = {J_AUTOM}, author = {Renjie Ma and Zhijian Hu and Rongni Yang and Ligang Wu}, doi = {10.1016/j.automatica.2025.112573}, journal = {Automatica}, month = {1}, pages = {112573}, shortjournal = {Automatica}, title = {Aperiodic-sampled neural network controllers with closed-loop stability verifications}, volume = {183}, year = {2026}, }
• (2026). Backstepping for partial differential equations: A survey. AUTOM, 183, 112572. (www)

  @article{ , abstract = {Systems modeled by partial differential equations (PDEs) are at least as ubiquitous as those by nature finite-dimensional and modeled by ordinary differential equations (ODEs). And yet, systematic and readily usable methodologies, for such a significant portion of real systems, have been historically scarce. Around the year 2000, the backstepping approach to PDE control began to offer not only a less abstract alternative to PDE control techniques replicating optimal and spectrum assignment techniques of the 1960s, but also enabled the methodologies of adaptive and nonlinear control, matured in the 1980s and 1990s, to be extended from ODEs to PDEs, allowing feedback synthesis for systems that are uncertain, nonlinear, and infinite-dimensional. The PDE backstepping literature has since grown to hundreds of papers and nearly a dozen books. This survey aims to facilitate the entry into this thriving area of overwhelming size and topical diversity. Designs of controllers and observers, for parabolic, hyperbolic, and other classes of PDEs, in one or more dimensions, with nonlinear, adaptive, sampled-data, and event-triggered extensions, are covered in the survey. The lifeblood of control are technology and physics. The survey places a particular emphasis on applications that have motivated the development of the theory and which have benefited from the theory and designs: flows, flexible structures, materials, thermal and chemically reacting dynamics, energy (from oil drilling to batteries and magnetic confinement fusion), and vehicles.}, archive = {J_AUTOM}, author = {Rafael Vazquez and Jean Auriol and Federico Bribiesca-Argomedo and Miroslav Krstic}, doi = {10.1016/j.automatica.2025.112572}, journal = {Automatica}, month = {1}, pages = {112572}, shortjournal = {Automatica}, title = {Backstepping for partial differential equations: A survey}, volume = {183}, year = {2026}, }
• (2026). Maximum principle for partial information non-zero sum stochastic differential games with mixed delays. AUTOM, 183, 112570. (www)

  @article{ , abstract = {This paper is concerned with one kind of partial information non-zero sum stochastic differential game with mixed delays. Both the state and control processes contain delays, where the former contains moving-average delay, discrete delay and noisy memory. We establish a necessary as well as two sufficient stochastic maximum principles for the game. As one of the main features of this research, a new kind of sufficient maximum principle is given, where the diffusion term can be controlled with non-convex control domains, and no second-order adjoint equation is needed. The theoretical results are applied to study two examples where the adjoint processes can be derived by two approaches and then the equilibrium points are obtained. This research generalizes those of stochastic optimal control problems.}, archive = {J_AUTOM}, author = {Pan Chen and Feng Zhang}, doi = {10.1016/j.automatica.2025.112570}, journal = {Automatica}, month = {1}, pages = {112570}, shortjournal = {Automatica}, title = {Maximum principle for partial information non-zero sum stochastic differential games with mixed delays}, volume = {183}, year = {2026}, }
• (2026). State estimation for lithium-ion batteries based on electrolyte–electrode PDE observers. AUTOM, 183, 112568. (www)

  @article{ , abstract = {Accurate information about the states of an electrochemical battery model facilitates a deeper understanding of battery behavior and enables performance enhancement. This paper first proposes backstepping Partial Differential Equation (PDE) observers for lithium concentration in the electrolyte phase within the negative electrode, positive electrode, and separator. Reverse sensitivity analysis is conducted to identify the most suitable measurable parameter for obtaining the electrolyte lithium concentration at the boundaries, which is used in the design of the electrolyte-phase observer. Subsequently, enhanced observers for the solid-phase lithium concentration in the negative and positive electrodes are developed. The proposed solid-phase observer enables more accurate State-of-Charge (SoC) estimation by leveraging the closed-loop electrolyte-phase observer. Simulations of the reverse sensitivity analysis and state observers are performed on a commercial cylindrical lithium iron phosphate ( LiFePO 4 ) cell to validate the effectiveness of the proposed approach.}, archive = {J_AUTOM}, author = {Sara Sepasiahooyi and Shu-Xia Tang}, doi = {10.1016/j.automatica.2025.112568}, journal = {Automatica}, month = {1}, pages = {112568}, shortjournal = {Automatica}, title = {State estimation for lithium-ion batteries based on electrolyte–electrode PDE observers}, volume = {183}, year = {2026}, }
• (2026). Interval-constraint multiagent systems: Global attractivity and structural stability of equilibria. AUTOM, 183, 112566. (www)

  @article{ , abstract = {This paper focuses on the dynamic behavior of an interval-constraint multiagent system. Each agent has a constraint interval that limits its potential consensus values, achieved by encoding a nonsmooth piecewise function into the agent. In addition, the underlying graphs considered are strongly connected. First, a dichotomy of equilibria is identified: either a unique non-consensus equilibrium point or multiple consensus points, depending on whether the intersection of the constraint intervals is empty or not. Then, the set of equilibria is proven to be a global attractor. Structural stability of such a system is also proven based on real-analysis methods, showing that the equilibria have continuous dependence on changes of the constraint intervals. Three running examples are used to illustrate the proposed results.}, archive = {J_AUTOM}, author = {Fengqiu Liu and Kuize Zhang and Yuhu Wu and Xiaoping Xue}, doi = {10.1016/j.automatica.2025.112566}, journal = {Automatica}, month = {1}, pages = {112566}, shortjournal = {Automatica}, title = {Interval-constraint multiagent systems: Global attractivity and structural stability of equilibria}, volume = {183}, year = {2026}, }
• (2026). Closed-loop data-enabled predictive control and its equivalence with closed-loop subspace predictive control. AUTOM, 183, 112556. (www)

  @article{ , abstract = {Factors like growing data availability and increasing system complexity have sparked interest in data-driven predictive control (DDPC) methods like Data-enabled Predictive Control (DeePC). However, closed-loop identification bias arises in the presence of noise, which reduces the effectiveness of obtained control policies. In this paper we propose Closed-loop Data-enabled Predictive Control (CL-DeePC), a framework that unifies different approaches to address this challenge. To this end, CL-DeePC incorporates instrumental variables (IVs) to synthesize and sequentially apply consistent single or multi-step-ahead predictors. Furthermore, a computationally efficient CL-DeePC implementation is developed that reveals an equivalence with Closed-loop Subspace Predictive Control (CL-SPC). Time marching simulations of DeePC and CL-DeePC are conducted using Hankel matrices of past data that are updated at every time step to induce potentially troublesome closed-loop correlations between inputs and noise. Compared to DeePC, CL-DeePC simulations demonstrate superior reference tracking, with a sensitivity study finding a 48% lower susceptibility to noise-induced reference tracking performance degradation.}, archive = {J_AUTOM}, author = {Rogier Dinkla and Tom Oomen and Sebastiaan Paul Mulders and Jan-Willem van Wingerden}, doi = {10.1016/j.automatica.2025.112556}, journal = {Automatica}, month = {1}, pages = {112556}, shortjournal = {Automatica}, title = {Closed-loop data-enabled predictive control and its equivalence with closed-loop subspace predictive control}, volume = {183}, year = {2026}, }
• (2026). Byzantine-resilient federated online learning for gaussian process regression. AUTOM, 183, 112554. (www)

  @article{ , abstract = {In this paper, we study Byzantine-resilient federated online learning for Gaussian process regression (GPR). We develop a Byzantine-resilient federated GPR algorithm that allows a cloud and a group of agents to collaboratively learn a latent function and improve the learning performances where some agents exhibit Byzantine failures, i.e., arbitrary and potentially adversarial behavior. Each agent-based local GPR sends potentially compromised local predictions to the cloud, and the cloud-based aggregated GPR computes a global model by a Byzantine-resilient product of experts aggregation rule. Then the cloud broadcasts the current global model to all the agents. Agent-based fused GPR refines local predictions by fusing the received global model with that of the agent-based local GPR. Moreover, we quantify the learning accuracy improvements of the agent-based fused GPR over the agent-based local GPR. Experiments on a toy example and two medium-scale real-world datasets are conducted to demonstrate the performances of the proposed algorithm.}, archive = {J_AUTOM}, author = {Xu Zhang and Zhenyuan Yuan and Minghui Zhu}, doi = {10.1016/j.automatica.2025.112554}, journal = {Automatica}, month = {1}, pages = {112554}, shortjournal = {Automatica}, title = {Byzantine-resilient federated online learning for gaussian process regression}, volume = {183}, year = {2026}, }
• (2026). Observer-based asymptotic active fault diagnosis: Separate and joint design of observer gain and input. AUTOM, 183, 112548. (www)

  @article{ , abstract = {This paper proposes observer gain and input design methods for observer-based asymptotic active fault diagnosis, which are based on a newly-defined notion named the excluding degree of the origin from a zonotope. Using the excluding degree, a quantitative specification is obtained to characterize the performance of set-based robust fault diagnosis. Furthermore, a separate gain design method and a joint gain and input design method are proposed, respectively. This is the first work to achieve a joint observer gain and input design for set-based active fault diagnosis. Compared with the existing methods that design gains and input separately, the proposed joint gain and input design method has advantages to exploit the fault diagnosis potential of observer-based schemes. Finally, several examples are used to illustrate the effectiveness of the proposed methods.}, archive = {J_AUTOM}, author = {Feng Xu and Yiming Wan and Ye Wang and Vicenç Puig}, doi = {10.1016/j.automatica.2025.112548}, journal = {Automatica}, month = {1}, pages = {112548}, shortjournal = {Automatica}, title = {Observer-based asymptotic active fault diagnosis: Separate and joint design of observer gain and input}, volume = {183}, year = {2026}, }
• (2026). Controller synthesis from noisy-input noisy-output data. AUTOM, 183, 112545. (www)

  @article{ , abstract = {We consider the problem of synthesizing a dynamic output-feedback controller for a linear system, using solely input–output data corrupted by measurement noise. To handle input–output data, an auxiliary representation of the original system is utilized. By exploiting the structure of the auxiliary system, we design a controller that robustly stabilizes all possible systems consistent with data. Notably, we also provide a novel solution to extend the results to generic multi-input multi-output systems. The findings are illustrated by numerical examples.}, archive = {J_AUTOM}, author = {Lidong Li and Andrea Bisoffi and Claudio De Persis and Nima Monshizadeh}, doi = {10.1016/j.automatica.2025.112545}, journal = {Automatica}, month = {1}, pages = {112545}, shortjournal = {Automatica}, title = {Controller synthesis from noisy-input noisy-output data}, volume = {183}, year = {2026}, }
• (2026). Accuracy bounds for the simulation of a class of continuous-time nonlinear models. AUTOM, 183, 112543. (www)

  @article{ , abstract = {Real-world dynamic systems evolve in the continuous-time world, while their models are simulated in the digital world using discrete-time numerical simulation algorithms. Such simulation is essential for a variety of system and control problems such as system identification and performance analysis of (control) systems. Ideally, the simulated model response should be identical to the system response. However, this is typically not the case in practice, even when the effects of unmodelled dynamics and parametric uncertainty are excluded. Even in that scenario, a mismatch exists between the response of the system and the model due to the interface between the physical world and the digital computer, unknown disturbances, and simulation inaccuracies. For the class of continuous-time, nonlinear Lur’e-type systems, this paper analyses the mismatch between the steady-state system response and the steady-state model response computed using the so-called mixed time–frequency algorithm. Firstly, a bound on the mismatch between the steady-state system response and the computed steady-state model response based on continuous-time signals is derived. Secondly, a bound for the same mismatch is derived for a sampled version of the signals. The bounds are further decomposed into several components, each given an interpretation that can be used to reduce the bounds on the mismatch. In a numerical case study, we show that reducing the bounds also reduces the actual mismatch.}, archive = {J_AUTOM}, author = {Fahim Shakib and Johan Schoukens and Alexander Yu. Pogromsky and Alexey Pavlov and Nathan van de Wouw}, doi = {10.1016/j.automatica.2025.112543}, journal = {Automatica}, month = {1}, pages = {112543}, shortjournal = {Automatica}, title = {Accuracy bounds for the simulation of a class of continuous-time nonlinear models}, volume = {183}, year = {2026}, }
• (2026). Revisiting lossless convexification: Theoretical guarantees for discrete-time optimal control problems. AUTOM, 183, 112537. (www)

  @article{ , abstract = {Lossless Convexification (LCvx) is a modeling approach that transforms a class of nonconvex optimal control problems, where nonconvexity primarily arises from control constraints, into convex problems through convex relaxations. These convex problems can be solved using polynomial-time numerical methods after discretization, which converts the original infinite-dimensional problem into a finite-dimensional one. However, existing LCvx theory is limited to continuous-time optimal control problems, as the equivalence between the relaxed convex problem and the original nonconvex problem holds only in continuous-time. This paper extends LCvx theory to discrete-time optimal control problems by classifying them into normal and long-horizon cases. For normal cases, after an arbitrarily small perturbation to the system dynamics (recursive equality constraints), applying the existing LCvx method to discrete-time problems results in optimal controls that meet the original nonconvex constraints at all but no more than n x − 1 temporal grid points, where n x is the state dimension. For long-horizon cases, the existing LCvx method fails, but we resolve this issue by integrating it with a bisection search, leveraging the continuity of the value function from the relaxed convex problem to achieve similar results as in normal cases. This paper strengthens the theoretical foundation of LCvx, extending the applicability of LCvx theory to discrete-time optimal control problems.}, archive = {J_AUTOM}, author = {Dayou Luo and Kazuya Echigo and Behçet Açıkmeşe}, doi = {10.1016/j.automatica.2025.112537}, journal = {Automatica}, month = {1}, pages = {112537}, shortjournal = {Automatica}, title = {Revisiting lossless convexification: Theoretical guarantees for discrete-time optimal control problems}, volume = {183}, year = {2026}, }
• (2026). Federated cubic regularized newton learning with sparsification-amplified differential privacy. AUTOM, 183, 112531. (www)

  @article{ , abstract = {This paper explores the cubic-regularized Newton method within a federated learning framework while addressing two major concerns: privacy leakage and communication bottlenecks. We propose the Differentially Private Federated Cubic Regularized Newton (DP-FCRN) algorithm, which leverages second-order techniques to achieve lower iteration complexity than first-order methods. We incorporate noise perturbation during local computations to ensure privacy. Furthermore, we employ sparsification in uplink transmission, which not only reduces the communication costs but also amplifies the privacy guarantee. Specifically, this approach reduces the necessary noise intensity without compromising privacy protection. We analyze the convergence properties of our algorithm and establish the privacy guarantee. Finally, we validate the effectiveness of the proposed algorithm through experiments on a benchmark dataset.}, archive = {J_AUTOM}, author = {Wei Huo and Changxin Liu and Kemi Ding and Karl Henrik Johansson and Ling Shi}, doi = {10.1016/j.automatica.2025.112531}, journal = {Automatica}, month = {1}, pages = {112531}, shortjournal = {Automatica}, title = {Federated cubic regularized newton learning with sparsification-amplified differential privacy}, volume = {183}, year = {2026}, }
• (2026). A consensus kalman filter on l2 spaces. AUTOM, 183, 112530. (www)

  @article{ , abstract = {We study the estimation problem of infinite dimensional discrete-time stochastic linear systems with finite dimensional measurements on sensor networks modeled by connected undirected graphs. The framework encompasses discretized PDEs with sampled measurements. A new scheme of distributed consensus on measurements is extended to systems evolving in L 2 spaces in order to limit the information exchange to finite-dimensional vectors. We show that, in analogy to the finite-dimensional case, at each node the variance of the estimation error tends to the one of the centralized Kalman filter for systems is L 2 when the number of consensus steps increases.}, archive = {J_AUTOM}, author = {Stefano Battilotti and Alessandro Borri and Filippo Cacace and Massimiliano d’Angelo and Alfredo Germani}, doi = {10.1016/j.automatica.2025.112530}, journal = {Automatica}, month = {1}, pages = {112530}, shortjournal = {Automatica}, title = {A consensus kalman filter on l2 spaces}, volume = {183}, year = {2026}, }
• (2026). Fixed-time learning for safe time-critical verification using reachability analysis. AUTOM, 183, 112528. (www)

  @article{ , abstract = {In this paper, we address a safe time-critical control problem using reachability analysis and design a reinforcement learning-based mechanism for learning online and in fixed-time the solution to the safe time-critical control problem. Safety is ensured by determining a set of states for which there exists an admissible control law generating a system trajectory that does not reach a set of forbidden states at a user-prescribed time instant. Specifically, we cast our safe time-critical problem as a Mayer optimal feedback control problem whose solution satisfies the Hamilton–Jacobi–Bellman (HJB) equation and characterizes the set of safe states. Since the HJB equation is generally difficult to solve, we develop an online critic-only reinforcement learning-based algorithm for simultaneously learning the solution to the HJB equation and the safe set in a fixed time. In particular, we introduce a non-Lipschitz experience replay-based learning law utilizing recorded and current data for updating the critic weights to learn the value function and the safe set. The non-Lipschitz property of the dynamics gives rise to fixed-time convergence, whereas the experience replay-based approach eliminates the need to satisfy the persistence of excitation condition provided that a recorded data set is sufficiently rich. Simulation results illustrate the efficacy of the proposed approach to the problem of fixed-wing unmanned aerial vehicle collision avoidance.}, archive = {J_AUTOM}, author = {Nick-Marios T. Kokolakis and Kyriakos G. Vamvoudakis and Wassim M. Haddad}, doi = {10.1016/j.automatica.2025.112528}, journal = {Automatica}, month = {1}, pages = {112528}, shortjournal = {Automatica}, title = {Fixed-time learning for safe time-critical verification using reachability analysis}, volume = {183}, year = {2026}, }
• (2026). Orchestrating on-board sensors for global hybrid robust stabilization of unicycles. AUTOM, 183, 112502. (www)

  @article{ , abstract = {We consider mobile robots described through unicycle dynamics equipped with on-board range sensors and cameras, one facing forward and one facing backward, providing measurements of the distance and misalignment to a target. We propose a hybrid control law combining the two on-board measurements and discuss stability results for the closed-loop expressed in the on-board camera-based coordinates, using Lyapunov-based arguments. We prove robustness of the stability properties to uncertainties affecting the sensors and external perturbations acting on the robot. The results are illustrated via simulations.}, archive = {J_AUTOM}, author = {Riccardo Ballaben and Alessandro Astolfi and Philipp Braun and Luca Zaccarian}, doi = {10.1016/j.automatica.2025.112502}, journal = {Automatica}, month = {1}, pages = {112502}, shortjournal = {Automatica}, title = {Orchestrating on-board sensors for global hybrid robust stabilization of unicycles}, volume = {183}, year = {2026}, }

CMA - 24

• (2025). A data-driven algorithm for solving image despeckling PDE model using physics-informed ConvNet. CMA, 200, 202-227. (www)

  @article{ , abstract = {In this study, we propose a new data-driven algorithm for the Perona-Malik image despeckling problem. The advantage of the proposed algorithm over neural network-based methods is that it does not need any noisy and clean image data pair for training. The proposed algorithm is implemented using a three-dimensional convolution neural network (ConvNet) architecture. We compare its output with results obtained from several existing methods, including the operator splitting RBF collocation method, the finite difference method (FDM), and physics-informed neural networks (PINNs). To evaluate the performance of the proposed algorithm, simulations are carried out using grayscale images that have been artificially corrupted with different levels of speckle noise. Using the peak signal to noise ratio (PSNR) and structural similarity index measure (SSIM) as the evaluation metric, we observed that the proposed algorithm outperforms these existing methods, demonstrating superior image quality with the same numerical scheme and the same discretization. To the best of our knowledge, this work represents the first application of physics-inspired convolutional neural network for PDE-based image despeckling model.}, archive = {J_CMA}, author = {Haridarshan Kumar and Sanjeev Kumar}, doi = {10.1016/j.camwa.2025.09.022}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {202-227}, shortjournal = {Comput. Meth. Appl.}, title = {A data-driven algorithm for solving image despeckling PDE model using physics-informed ConvNet}, volume = {200}, year = {2025}, }
• (2025). An easy to implement numerical framework for a cancer invasion mathematical model with two distinct cancer sub-populations. CMA, 200, 180-201. (www)

  @article{ , abstract = {This work presents a finite element scheme for solving a model that involves sub-populations of cancer cells. The model is formulated by four coupled partial differential equations, which represent the evolution of tumor growth, the density of cancer cell sub-populations arising from mutations, the density of the extracellular matrix (ECM), and the concentration of matrix-degrading enzymes (MDE). A semi-implicit backward Euler finite element framework has been developed for this model. Unconditional error estimates have been established for all variables, and the unconditional stability of the solutions has also been demonstrated. To validate the proposed numerical scheme, we have performed numerical simulations, including an assessment of the convergence rate and comparisons between the numerical solutions and analytical solutions.}, archive = {J_CMA}, author = {Yadhavan Karuppusamy and Lingeshwaran Shangerganesh and Sally Mohammed Farghaly Abdelaliem and A.S. Hendy}, doi = {10.1016/j.camwa.2025.09.011}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {180-201}, shortjournal = {Comput. Meth. Appl.}, title = {An easy to implement numerical framework for a cancer invasion mathematical model with two distinct cancer sub-populations}, volume = {200}, year = {2025}, }
• (2025). Stabilize physics-informed neural networks for stiff differential equations: Re-spacing layer. CMA, 200, 167-179. (www)

  @article{ , abstract = {Approximating the solution of stiff differential equations, which exhibit abrupt changes in certain regions, using physics-informed neural networks (PINNs) is challenging. Typically, training PINNs involves using a larger number of samples concentrated around regions of rapid changes to resolve the sharp gradients. However, this strategy leads to data imbalance, resulting in slower convergence and reduced solution quality. Here, we propose Re-spacing layer (RS-layer) to mitigate these challenges. RS-layer is a pre-trained encoding layer designed to map the skewed distribution of sampling points onto a uniform distribution, maintaining the desirable statistical properties of the input data for effective PINN training. We demonstrate that RS-layer improves PINN training by regularizing the solution gradient in the transformed space. The efficacy of our method is validated through numerical experiments on one-dimensional singularly perturbed equations, the ROBER problem, and the Akzo Nobel problem. Our results show that RS-layer not only accelerates convergence, but also enhances accuracy.}, archive = {J_CMA}, author = {Eunsuh Kim and Heejae Kwon and Sungha Cho and Kyongmin Yeo and Minseok Choi}, doi = {10.1016/j.camwa.2025.09.014}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {167-179}, shortjournal = {Comput. Meth. Appl.}, title = {Stabilize physics-informed neural networks for stiff differential equations: Re-spacing layer}, volume = {200}, year = {2025}, }
• (2025). Hp-version discontinuous galerkin time-stepping schemes for diffusive-viscous wave equation. CMA, 200, 145-166. (www)

  @article{ , abstract = {This work introduces a fully discrete scheme for the diffusion-viscous wave equation (DVWe) in the second-order formulation, combining hp -DG time-stepping schemes with conforming finite element methods (FEM). Two major theoretical contributions are presented: (1) hp -version a priori error estimates in both the energy-norm and DG-norm, which are optimal in the spatial mesh size h , temporal step size τ , and temporal polynomial order q , yet suboptimal by one order in the spatial polynomial order p . Furthermore, for solutions exhibiting weak singularities in time, exponential convergence in terms of the total number of temporal degrees of freedom is proven using the hp -refinement strategy. (2) An energy decay estimate that offers explicit bounds involving the model parameters, discretization parameters, and the Poincaré inequality constant. A series of numerical experiments are presented to validate the practical performance of the proposed approach.}, archive = {J_CMA}, author = {Min Zhang and Zhaonan Dong and Wenjing Yan}, doi = {10.1016/j.camwa.2025.09.021}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {145-166}, shortjournal = {Comput. Meth. Appl.}, title = {Hp-version discontinuous galerkin time-stepping schemes for diffusive-viscous wave equation}, volume = {200}, year = {2025}, }
• (2025). Approximation of unknown sources in a time fractional PDE by the optimal ones and their reconstruction. CMA, 200, 117-144. (www)

  @article{ , abstract = {In this paper, our focus is on studying a geometric inverse source problem that is governed by two-dimensional time-fractional subdiffusion. The problem involves determining the shape and location of the unknown source's geometrical support from boundary measurements of its associated potential. Firstly, we prove the uniqueness of the inverse problem. In the second phase, we propose a novel reconstruction method that utilizes the coupled complex boundary method (CCBM) to solve the identification problem. The main idea of this method is to approximate the overdetermined problem to a complex boundary value problem with a complex Robin boundary condition coupling the Dirichlet and Neumann boundary conditions. Next, we utilize the imaginary part of the solution throughout the domain to construct a shape cost function, which we then minimize with respect to ball-shaped sources by using a Newton-type topological derivative method to reconstruct the geometrical support of the unknown source.}, archive = {J_CMA}, author = {Mourad Hrizi and Ravi Prakash and Antonio André Novotny}, doi = {10.1016/j.camwa.2025.09.020}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {117-144}, shortjournal = {Comput. Meth. Appl.}, title = {Approximation of unknown sources in a time fractional PDE by the optimal ones and their reconstruction}, volume = {200}, year = {2025}, }
• (2025). An averaged l1 ADI compact difference scheme for the three-dimensional time-fractional mobile/immobile transport equation with weakly singular solutions. CMA, 200, 102-116. (www)

  @article{ , abstract = {In this paper, a three-dimensional (3D) time-fractional mobile/immobile (MIM) transport equation, which incorporates the Caputo time-fractional derivative of order α ∈ ( 0 , 1 ) , is taken into consideration. The space derivatives are discretized using the compact finite difference approximation, and the Caputo time-fractional derivative is estimated by employing the averaged L1 formula. Combining with corresponding alternating direction implicit (ADI) algorithms, the overall computational cost is reduced significantly. Using the discrete energy analysis methodology, we demonstrate that the suggested method possesses temporal second-order convergence and spatial fourth-order convergence under the regularity assumption. Numerical experiments demonstrate that ADI techniques is effective in computing 3D problems.}, archive = {J_CMA}, author = {Kai Liu and Haixiang Zhang and Xuehua Yang}, doi = {10.1016/j.camwa.2025.09.019}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {102-116}, shortjournal = {Comput. Meth. Appl.}, title = {An averaged l1 ADI compact difference scheme for the three-dimensional time-fractional mobile/immobile transport equation with weakly singular solutions}, volume = {200}, year = {2025}, }
• (2025). Pricing american options with exogenous and endogenous transaction costs. CMA, 200, 85-101. (www)

  @article{ , abstract = {We study an American option pricing problem with liquidity risks and transaction fees. As endogenous transaction costs, liquidity risks of the underlying asset are modeled by a mean-reverting process. Transaction fees are exogenous transaction costs and are assumed to be proportional to the trading amount, with the long-run liquidity level depending on the proportional transaction costs rate. Two nonlinear partial differential equations are established to characterize the option values for the holder and the writer, respectively. To illustrate the impact of these transaction costs on option prices and optimal exercise prices, we apply the alternating direction implicit method to solve the linear complementarity problem numerically. Finally, we conduct model calibration from market data via maximum likelihood estimation, and find that our model incorporating liquidity risks outperforms the Leland model significantly.}, archive = {J_CMA}, author = {Dong Yan and Xin-Jie Huang and Guiyuan Ma and Xin-Jiang He}, doi = {10.1016/j.camwa.2025.09.008}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {85-101}, shortjournal = {Comput. Meth. Appl.}, title = {Pricing american options with exogenous and endogenous transaction costs}, volume = {200}, year = {2025}, }
• (2025). A space-time discontinuous petrov-galerkin finite element formulation for a modified schrödinger equation for laser pulse propagation in waveguides. CMA, 200, 67-84. (www)

  @article{ , abstract = {In this article, we propose a modified nonlinear Schrödinger equation for modeling pulse propagation in optical waveguides. The proposed model bifurcates into a system of elliptic and hyperbolic equations depending on waveguide parameters. The proposed model leads to a stable first-order system of equations, distinguishing itself from the canonical nonlinear Schrödinger equation. We have employed the space-time discontinuous Petrov-Galerkin finite element method to discretize the first-order system of equations. We present a stability analysis for both the elliptic and hyperbolic systems of equations and demonstrate the stability of the proposed model through several numerical examples on space-time meshes.}, archive = {J_CMA}, author = {A. Chakraborty and J. Muñoz-Matute and L. Demkowicz and J. Grosek}, doi = {10.1016/j.camwa.2025.09.004}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {67-84}, shortjournal = {Comput. Meth. Appl.}, title = {A space-time discontinuous petrov-galerkin finite element formulation for a modified schrödinger equation for laser pulse propagation in waveguides}, volume = {200}, year = {2025}, }
• (2025). Convergence analysis of an energy-stable linearized virtual element method for the strongly damped klein-gordon equation. CMA, 200, 49-66. (www)

  @article{ , abstract = {In this paper, we propose and analyze an efficient, linearized, fully discrete scheme for the nonlinear, strongly damped Klein-Gordon equation on polygonal meshes. The numerical scheme uses a conforming virtual element method for spatial discretization and a modified leapfrog (central finite difference) scheme for time discretization, with the nonlinear term | u | p − 1 u is treated semi-implicitly. We first prove that the proposed scheme is energy dissipative in the sense of discrete energy, and then the stability of the numerical solution in the H 1 -norm is established using mathematical induction, which plays an important role in handling the nonlinear term. By applying the boundedness of the numerical solution and the Sobolev embedding inequality, we derive the optimal H 1 error estimate of order O ( h k + τ 2 ) without imposing any ratio restrictions between the time step τ and the mesh size h . Additionally, we remark that the leapfrog virtual element scheme can be applied to some more complex nonlinear damped wave equations. Finally, some numerical examples are provided to confirm the theoretical results.}, archive = {J_CMA}, author = {Zhixin Liu and Minghui Song and Yuhang Zhang}, doi = {10.1016/j.camwa.2025.09.002}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {49-66}, shortjournal = {Comput. Meth. Appl.}, title = {Convergence analysis of an energy-stable linearized virtual element method for the strongly damped klein-gordon equation}, volume = {200}, year = {2025}, }
• (2025). A posteriori error estimate of the discontinuous galerkin method with lagrange multiplier for elliptic problems. CMA, 200, 38-48. (www)

  @article{ , abstract = {This study aims to derive and analyze an a posteriori error estimator for the solution of the discontinuous Galerkin method with Lagrange multiplier (DGLM) for the elliptic problems with nonhomogeneous Dirichlet boundary condition u = g for g in H 1 / 2 ( ∂ Ω ) . A general version of the DGLM method is derived. Strong stability of the solution of the DGLM method is proved. Edgewise iterative scheme for the general DGLM method is described.}, archive = {J_CMA}, author = {Mi-Young Kim}, doi = {10.1016/j.camwa.2025.09.005}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {38-48}, shortjournal = {Comput. Meth. Appl.}, title = {A posteriori error estimate of the discontinuous galerkin method with lagrange multiplier for elliptic problems}, volume = {200}, year = {2025}, }
• (2025). Stability and convergence analysis of mixed finite element approximations for a biot-brinkman model. CMA, 200, 22-37. (www)

  @article{ , abstract = {Many multiphysics processes of fluid-solid interaction within a porous medium can be described by the Biot-Brinkman model to account for the effects of viscosity in fluid flow. By introducing the auxiliary variables, we can transform the original problem into two generalized Stokes equations. The generalized Stokes equations incorporate a built-in mechanism to circumvent the Poisson locking for the continuous Galerkin method. Subsequently, we establish an energy law and provide a priori estimates for the reformulated problem. Well-posedness is demonstrated using the standard Galerkin method in conjunction with a compactness argument. After that, we develop stable mixed finite element algorithms for the reformulated problem. Influenced by Lamé constant λ , we design three finite element pairs for the proposed algorithms and present the corresponding error estimates. Numerical tests are conducted to validate the theoretical results.}, archive = {J_CMA}, author = {Wenlong He and Jiwei Zhang}, doi = {10.1016/j.camwa.2025.09.006}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {22-37}, shortjournal = {Comput. Meth. Appl.}, title = {Stability and convergence analysis of mixed finite element approximations for a biot-brinkman model}, volume = {200}, year = {2025}, }
• (2025). Physics-informed neural network for option pricing weather derivatives model. CMA, 200, 1-21. (www)

  @article{ , abstract = {Weather derivatives are financial tools that use a weather index as the underlying asset to provide protection against non-catastrophic weather events. In this article, we propose a physics-informed neural network (PINN) approach for pricing weather derivatives associated with two standard processes: the Ornstein-Uhlenbeck process and the Ornstein-Uhlenbeck process with jump-diffusions. PINNs are a scientific machine learning method specifically designed to address problems related to partial differential equations (PDEs). To apply the PINN technique for jump-diffusion, we convert the partial integro-differential equation into a PDE using integral discretization. We randomly select training data points within the domain and utilize the transformed PDE along with the initial and boundary conditions to construct the loss function. For the neurons in the hidden layer, we employ the hyperbolic tangent function (tanh) as the activation function. The weights of the network connection are optimized using the L-BFGS algorithm. We will conduct numerical experiments to evaluate the efficiency of the proposed technique. Additionally, we compare our method with conventional numerical approaches to show that our technique serves as an effective alternative to existing pricing methods for weather derivatives. Finally, we will examine a real-world case study where the model's parameters are determined using precipitation data.}, archive = {J_CMA}, author = {Saurabh Bansal and Pradanya Boro and Srinivasan Natesan}, doi = {10.1016/j.camwa.2025.09.001}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {1-21}, shortjournal = {Comput. Meth. Appl.}, title = {Physics-informed neural network for option pricing weather derivatives model}, volume = {200}, year = {2025}, }
• (2025). A bound-preserving scheme for the Allen–Cahn equation. CMA, 199, 225-241. (www)

  @article{ , abstract = {We propose and analyze a bound-preserving scheme for the Allen–Cahn equation. The key idea is to apply a bound-preserving nonlinear stabilization technique to the implicit Euler time-stepping method coupled with the continuous finite element method. To our best knowledge, this is the first scheme which theoretically preserves the maximum principle and has an error estimate that is optimal in the L 2 ( H 1 ) -seminorm and with a polynomial dependence on ϵ − 1 at the same time. The proof of the error estimate combines a nonlinear Ritz projection together with a special Grönwall inequality. Numerical experiments are conducted to compare the performance of our scheme with a bound-preserving operator-splitting scheme.}, archive = {J_CMA}, author = {Zhaonan Dong and Alexandre Ern and Zuodong Wang}, doi = {10.1016/j.camwa.2025.09.025}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {225-241}, shortjournal = {Comput. Meth. Appl.}, title = {A bound-preserving scheme for the Allen–Cahn equation}, volume = {199}, year = {2025}, }
• (2025). A local weak form of the generalized finite difference method (GFDM) with control volume in heat conduction problems. CMA, 199, 203-224. (www)

  @article{ , abstract = {An explicit local weak form of the Generalized Finite Difference Method (GFDM) is developed for solving heat conduction problems by incorporating the concept of control volumes from the Finite Volume Method (FVM). The control volume is introduced as the local integral domain in the local weak form, where the Heaviside function is employed as the weight function, similar to traditional FVM. However, instead of the cell-centered control volume used in FVM, a vertex-centered control volume is applied for integration. The trial function is constructed using a Generalized Finite Difference approximation based on a Taylor expansion. By applying the divergence theorem, the integral of the governing equation over the control volume is transformed into a boundary integral, thereby reducing the continuity requirements for both the trial function and thermal conductivity. Several numerical examples demonstrate the proposed method's accuracy, stability, and convergence. Additionally, it offers local integration scheme without requiring any background grid.}, archive = {J_CMA}, author = {Tao Zhang and Xiaofeng Zhou}, doi = {10.1016/j.camwa.2025.09.015}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {203-224}, shortjournal = {Comput. Meth. Appl.}, title = {A local weak form of the generalized finite difference method (GFDM) with control volume in heat conduction problems}, volume = {199}, year = {2025}, }
• (2025). Hybrid active-passive control of multiphase composite shells through high-order chebyshev frequency response modeling. CMA, 199, 184-202. (www)

  @article{ , abstract = {This paper presents a high-efficiency approach for controlled frequency response analysis of laminated multi-phase composite shells, employing high-order Chebyshev finite elements. A key feature of the composite material used in this study is its three-phase composition, consisting of a polymer matrix, carbon fiber, and carbon nanotube nanofillers embedded within the matrix phase. The inclusion of carbon nanotube nanofillers significantly enhances the mechanical and dynamic properties of the overall material. The proposed approach addresses shear/membrane locking and the computational expense of discrete layerwise models, ensuring both efficiency and precision in analyzing complex composite shells. By using high-order shape functions derived from Chebyshev polynomials, this approach achieves rapid convergence without sacrificing accuracy. The developed numerical model captures the frequency response of laminated shells across varying geometric configurations, while incorporating the effects of two-layer control patches that introduce active or passive damping. Additionally, the model accounts for the layerwise effect of the composite, allowing for accurate prediction of structural behavior under excited loads. Numerical validation confirms the robustness and versatility of the high-order Chebyshev finite elements, effectively overcoming common computational challenges such as locking and spurious modes, while providing highly accurate and reliable frequency response predictions for practical engineering applications. This study highlights the potential of using advanced numerical techniques in combination with innovative composite materials for improved dynamic performance in aerospace, automotive, and civil engineering applications.}, archive = {J_CMA}, author = {Duy-Khuong Ly and Huy-Cuong Vu-Do and Trung-Hau Dang and T. Nguyen-Thoi}, doi = {10.1016/j.camwa.2025.09.012}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {184-202}, shortjournal = {Comput. Meth. Appl.}, title = {Hybrid active-passive control of multiphase composite shells through high-order chebyshev frequency response modeling}, volume = {199}, year = {2025}, }
• (2025). Bound-preserving physics-informed neural networks for steady-state convection-diffusion-reaction problems. CMA, 199, 167-183. (www)

  @article{ , abstract = {Numerical approximations of solutions of convection-diffusion-reaction problems should take only physically admissible values. Provided that bounds for the admissible values are known, this paper presents several approaches within physics-informed neural networks (PINNs) and hp -variational PINNs ( hp -VPINNs) to preserve these bounds for convection-dominated problems. These approaches comprise the inclusion of the requirement for bound preservation in the cost functional, a simple cut-off strategy for the unphysical values, and two methods that enforce bound preservation via the activation function of the output layer of the neural network. Numerical simulations are performed for convection-dominated problems defined in two-dimensional domains. A variety of choices for several hyperparameters is explored. Enforcing bound preservation with the sine activation function in the output layer turned out to be superior to all other methods with respect to the accuracy of the computed solutions, and in particular, the results are much more accurate than those obtained with the standard PINNs and hp -VPINNs.}, archive = {J_CMA}, author = {Marina Matthaiou and Volker John and Marwa Zainelabdeen}, doi = {10.1016/j.camwa.2025.09.009}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {167-183}, shortjournal = {Comput. Meth. Appl.}, title = {Bound-preserving physics-informed neural networks for steady-state convection-diffusion-reaction problems}, volume = {199}, year = {2025}, }
• (2025). A bond-based nonlocal anisotropic diffusion model with variable matrix-valued coefficients and its asymptotically compatible meshfree discretization. CMA, 199, 148-166. (www)

  @article{ , abstract = {We propose a bond-based nonlocal anisotropic diffusion model with variable matrix-valued coefficients. In previous studies, the non-ordinary state-based nonlocal diffusion model [37] has been used effectively for simulating anisotropic diffusion. However, it encounters challenges such as high computational costs, complexities in implementing boundary conditions, and numerical oscillation of zero-energy mode. In this paper, we propose a novel bond-based, nonlocal anisotropic diffusion model, and the key idea is that incorporates a nonlocal operator via a kernel function, integrating matrix-valued diffusion coefficients. The influence region of our model consists of two parts: an elliptical region determined by the variable diffusion coefficient at a material point and an irregular region shaped by the coefficient at neighboring points. Furthermore, we confirm the well-posedness of the proposed model and deduce various properties, such as weak convergence and mass conservation. For computational implementation, we introduce a meshfree method that is shown to be asymptotically compatible and relies on the quadrature rule, which is compatible with the proposed nonlocal diffusion model and can effectively solve the model. To evaluate the precision and efficiency of the model, we performed comprehensive numerical experiments in both two and three dimensions. We have also confirmed the discrete maximum principle through experimental validation simultaneously.}, archive = {J_CMA}, author = {Xiaofang Wang and Hao Tian}, doi = {10.1016/j.camwa.2025.09.018}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {148-166}, shortjournal = {Comput. Meth. Appl.}, title = {A bond-based nonlocal anisotropic diffusion model with variable matrix-valued coefficients and its asymptotically compatible meshfree discretization}, volume = {199}, year = {2025}, }
• (2025). Shape derivative of the laplacian eigenvalue problem. CMA, 199, 127-147. (www)

  @article{ , abstract = {The Laplacian eigenvalue problem with two different densities is investigated. By the squeeze theorem, the shape derivative of the least eigenvalue on the interface of two different density subdomains is derived. As an application, the minimization of the least eigenvalue with area constraint is considered. The shape derivative of the objective functional is applied as the velocity for the level set method to involve the interface. The numerical results validate that the proposed method is effective to capture the final optimized distribution of two different densities.}, archive = {J_CMA}, author = {Zhengfang Zhang and Lulu Guo and Xiangjing Gao and Weifeng Chen and Xiaoliang Cheng}, doi = {10.1016/j.camwa.2025.09.013}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {127-147}, shortjournal = {Comput. Meth. Appl.}, title = {Shape derivative of the laplacian eigenvalue problem}, volume = {199}, year = {2025}, }
• (2025). Regular boundary element method for composite shear deformable plate and shell. CMA, 199, 106-126. (www)

  @article{ , abstract = {This paper presents a fundamental solution for a double-curvature simply supported shell, incorporating three concentrated forces and two bending moments. It introduces the reference domain concept and formulates fictitious load boundary integral equations using both constant and linear elements. These equations are developed in the Laplace transform domain for both static and dynamic problems. The key contribution of this study is the development of the Regular Boundary Element Method (RBEM) based on the new fundamental solution. The reference domain includes the real structure’s configuration, and a system of linear equations is established with fictitious forces and moments as unknowns. These equations are derived from traction and displacement boundary conditions. To obtain all physical values in the time domain, the Durbin’s Laplace inverse technique is applied. The accuracy and reliability of the proposed method are evaluated through four numerical examples, with results compared against exact solutions or the finite element method.}, archive = {J_CMA}, author = {W. Huang and X.B. Yan and J.X. Liao and L.K. Feng and P.H. Wen}, doi = {10.1016/j.camwa.2025.09.016}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {106-126}, shortjournal = {Comput. Meth. Appl.}, title = {Regular boundary element method for composite shear deformable plate and shell}, volume = {199}, year = {2025}, }
• (2025). Source point selection in the MFS using orthogonal matching pursuit: Two-dimensional elastic wave scattering by a rectangular cavity. CMA, 199, 80-105. (www)

  @article{ , abstract = {In this paper, we apply orthogonal matching pursuit (OMP) to the source point selection in the method of fundamental solutions (MFS) and discuss its effectiveness. The proposed method initially places an excess number of source points relative to the collocation points within the complementary domain of the analysis region and then selects the source points that efficiently reduce the residual of the system of equations. We apply the proposed method to two-dimensional elastic wave scattering by a rectangular cavity, a well-known challenging problem in MFS analysis. Numerical examples demonstrate the effectiveness of the proposed method by comparing its performance with the conventional MFS using truncated singular value decomposition (TSVD). The proposed method provides solutions that more accurately satisfy the traction-free boundary conditions compared to the conventional method, indicating its potential advantages.}, archive = {J_CMA}, author = {Akira Furukawa}, doi = {10.1016/j.camwa.2025.09.003}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {80-105}, shortjournal = {Comput. Meth. Appl.}, title = {Source point selection in the MFS using orthogonal matching pursuit: Two-dimensional elastic wave scattering by a rectangular cavity}, volume = {199}, year = {2025}, }
• (2025). A DOFs condensation based algorithm for solving saddle point systems in 2D contact computation. CMA, 199, 64-79. (www)

  @article{ , abstract = {In contact mechanics computation, the constraint conditions on the contact surfaces are typically enforced by the Lagrange multiplier method, resulting in a saddle point system. The mortar finite element method is usually employed to discretize the variational form on the meshed contact surfaces, yielding a large-scale discretized saddle point system. Due to the indefiniteness of the discretized system, it is a challenge to solve the saddle point algebraic system. For two-dimensional tied contact problem, we develop an efficient algorithm based on degree-of-freedom (DOF) condensation. In this approach, a DOFs elimination process is first performed by exploiting the tridiagonal structure of the mortar matrix. The reduced linear system, now smaller in scale and symmetric positive definite (SPD), is then solved using the preconditioned conjugate gradient (PCG) method. Numerical results demonstrate the effectiveness of the algorithm.}, archive = {J_CMA}, author = {Xiaoyu Duan and Zihan Wang and Hengbin An and Zeyao Mo}, doi = {10.1016/j.camwa.2025.08.032}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {64-79}, shortjournal = {Comput. Meth. Appl.}, title = {A DOFs condensation based algorithm for solving saddle point systems in 2D contact computation}, volume = {199}, year = {2025}, }
• (2025). Data-driven SFB solutions and parameters discovery for nonlinear schrödinger equation via time domain decomposition physics-informed neural networks. CMA, 199, 45-63. (www)

  @article{ , abstract = {In this paper, we integrate domain decomposition techniques into the classical physics-informed neural networks (PINNs) by introducing interface training points, and propose a time domain decomposition PINNs (TDD-PINNs) framework. This model is applied to investigate the dynamic behaviour of solitons on finite background (SFB) solutions and parameter discovery in the nonlinear Schrödinger equation (NLSE). The TDD-PINNs is employed to study various SFB solutions, including the Akhmediev breather, Peregrine soliton, Kuznetsov-Ma soliton, as well as second- and third-order rogue waves. Experimental results demonstrate that, compared to classical PINNs, the proposed TDD-PINNs significantly reduce training time and improve prediction accuracy by one to two orders of magnitude. For inverse problems, the TDD-PINNs algorithm can accurately identify unknown parameters in the NLSE, both under noisy and noise-free conditions, addressing the complete failure of classical PINNs in parameter identification for NLSE and demonstrating strong robustness.}, archive = {J_CMA}, author = {Jiaxin Chen and Biao Li and Manwai Yuen}, doi = {10.1016/j.camwa.2025.09.007}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {45-63}, shortjournal = {Comput. Meth. Appl.}, title = {Data-driven SFB solutions and parameters discovery for nonlinear schrödinger equation via time domain decomposition physics-informed neural networks}, volume = {199}, year = {2025}, }
• (2025). Mixed nitsche extended finite element method for solving three-dimensional h(curl)-elliptic interface problems. CMA, 199, 22-44. (www)

  @article{ , abstract = {In this paper, we introduce a Lagrange multiplier to relax the divergence-free constraint and propose a mixed Nitsche extended finite element method for solving three-dimensional H(curl)-elliptic interface problems. To ensure stability, we incorporate ghost penalty terms. By exploiting the commuting relationship of the de Rham complex, we derive an inf-sup stability result for the discrete bilinear form, which is uniform with respect to the mesh size, discontinuous parameters, and the interface position. Based on this, we establish the well-posedness of our method and demonstrate optimal error bounds in the discrete energy norm and L 2 norm. Finally, numerical experiments are presented to illustrate the theoretical results.}, archive = {J_CMA}, author = {Nan Wang and Hanyu Chu and Jinru Chen and Ying Cai}, doi = {10.1016/j.camwa.2025.08.031}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {22-44}, shortjournal = {Comput. Meth. Appl.}, title = {Mixed nitsche extended finite element method for solving three-dimensional h(curl)-elliptic interface problems}, volume = {199}, year = {2025}, }
• (2025). Convergence analysis of a fast ADI compact finite difference method for two-dimensional semi-linear time-fractional reaction-diffusion equations with weak initial singularity. CMA, 199, 1-21. (www)

  @article{ , abstract = {In this work, considering the solution's weak initial singularity, a rigorous error analysis of a finite difference method for simulating a two-dimensional semi-linear time-fractional reaction-diffusion equation (TFRDE) is presented. The recently introduced ADI method by Kumari and Roul (2024) [31] for solving a class of linear TFRDEs encounters with problematic mesh parameter adjustments and ignorance of the derivative bounds, potentially rendering the latest methodology deficient and erroneous. The present study aims to design a computationally efficient L1 ADI scheme for semi-linear TFRDEs and provide a comprehensive error analysis. To address intrinsically non-local characteristics of the solution, we employ sum-of-exponential approximation to the singular kernel of time-fractional derivative on a graded mesh with unequal time-steps that yield denser mesh near the initial point. As a result, we effectively mitigate the high storage and computational requirements and return the convergence point to its optimal state. The two spatial variables are treated with a fourth order compact finite difference operator. Moreover, an alternating direction implicit method is utilized to compute the solution of the derived two-dimensional system by splitting it into two separate one-dimensional problems. With the aid of local truncation error estimate and discrete fractional Grönwall inequality, the stability and convergence analysis of the scheme are carried out rigorously through the discrete energy approach. The numerical results corroborate the convergence analysis and highlight the computational efficacy of the numerical scheme. Numerical examples demonstrate the CPU performance of the fast compact ADI method, and presented comparisons distinctly showcases the effectiveness of the graded mesh enhancing convergence order to achieve optimal results.}, archive = {J_CMA}, author = {Priyanka and Sunil Kumar}, doi = {10.1016/j.camwa.2025.08.028}, journal = {Computers & Mathematics with Applications}, month = {12}, pages = {1-21}, shortjournal = {Comput. Meth. Appl.}, title = {Convergence analysis of a fast ADI compact finite difference method for two-dimensional semi-linear time-fractional reaction-diffusion equations with weak initial singularity}, volume = {199}, year = {2025}, }

CMAME - 13

• (2026). GPU-accelerated multi-phase, multi-resolution SPH method with ray tracing for laser powder bed fusion. CMAME, 448, 118423. (www)

  @article{ , abstract = {Powder-scale simulation of laser powder bed fusion (LPBF) is increasingly vital for understanding, predicting, and controlling metallurgical defects. However, the complex multi-physics and multi-material interactions involved, along with high computational demands, pose significant challenges. This study presents the first multiphase smoothed particle hydrodynamics (SPH) simulation framework for LPBF under both low and high evaporation regimes, incorporating a multi-resolution particle strategy and ray tracing (RT). An adaptive particle refinement (APR) method, compatible with multiphase SPH and optimized for GPU acceleration, is developed to enhance computational efficiency of the multiphase model. The RT model is also optimized and integrated with the APR-GPU architecture, further improving performance. A physics-based wetting force model is introduced, along with a novel method for improving normal vector accuracy near the contact line. The proposed framework is validated through three benchmark cases and applied to simulate LPBF processes. The results demonstrate that the framework achieves high accuracy and efficiency in resolving key LPBF phenomena, including melt pool dynamics and keyhole formation.}, archive = {J_CMAME}, author = {Yibo Ma and Zhilang Zhang and Christian Weißenfels and Minli Zhou and Lingxiao Ma and Xiaofei Tang and Moubin Liu}, doi = {10.1016/j.cma.2025.118423}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118423}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {GPU-accelerated multi-phase, multi-resolution SPH method with ray tracing for laser powder bed fusion}, volume = {448}, year = {2026}, }
• (2026). Sparse data assimilation for under-resolved large-eddy simulations. CMAME, 448, 118421. (www)

  @article{ , abstract = {The need for accurate and fast scale-resolving simulations of fluid flows, where turbulent dispersion is a crucial physical feature, is evident. Large-eddy simulations (LES) are computationally more affordable than direct numerical simulations, but their accuracy depends on sub-grid scale models and the quality of the computational mesh. In order to compensate related errors, a data assimilation approach for LES is devised in this work. The presented method is based on variational assimilation of sparse time-averaged velocity reference data. Working with the time-averaged LES momentum equation allows to employ a stationary discrete adjoint method. Therefore, a stationary corrective force in the unsteady LES momentum equation is iteratively updated within the gradient-based optimization framework in conjunction with the adjoint gradient. After data assimilation, corrected anisotropic Reynolds stresses are inferred from the stationary corrective force. Ultimately, this corrective force that acts on the mean velocity is replaced by a term that scales the velocity fluctuations through nudging of the corrected anisotropic Reynolds stresses. Efficacy of the proposed framework is demonstrated for turbulent flow over periodic hills and around a square cylinder. Coarse meshes are leveraged to further enhance the speed of the optimization procedure. Time- and spanwise-averaged velocity reference data from high-fidelity simulations is taken from the literature. Our results demonstrate that adjoint-based assimilation of averaged velocity enables the optimization of the mean flow, vortex shedding frequency (i. e., Strouhal number), and anisotropic Reynolds stresses. This highlights the superiority of scale-resolving simulations such as LES over simulations based on the (unsteady) Reynolds-averaged equations.}, archive = {J_CMAME}, author = {Justin Plogmann and Oliver Brenner and Patrick Jenny}, doi = {10.1016/j.cma.2025.118421}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118421}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Sparse data assimilation for under-resolved large-eddy simulations}, volume = {448}, year = {2026}, }
• (2026). DGTO: Derivable geodesics-coupled topology optimization for multi-axis 3D printing of continuous fiber-reinforced spatial structures. CMAME, 448, 118419. (www)

  @article{ , abstract = {Continuous fiber reinforced composites (CFRCs) are composite materials with exceptional mechanical properties to enhance structural mechanical performance. In comparison with traditional three-axis 3D printing (also referred to as 2.5D printing), multi-axis 3D printing simultaneously moves the nozzle and rotates the build platform during the printing process, making it particularly suited for fabricating spatial structures made of CFRCs due to better alignment between the fiber depositions and principal stress directions. In this research, we propose a Derivable Geodesics-coupled Topology Optimization (DGTO) method for design of CFRCs given the manufacturing scheme of multi-axis 3D printing. A prominent feature of DGTO is the introduction of two geodesic fields to achieve curved layer generation and continuous fiber path planning. The heat diffusion equation and Poisson equation are solved to produce the geodesic fields, and hence, all objective functions and constraints related to the slices and paths are derivable, making them perfectly suitable to be integrated with topology optimization. Hence, the proposed method concurrently optimizes the density field and the auxiliary geodesic fields, simultaneously tuning the material distribution and spatial fiber distribution, thereby attaining optimal performance while fulfilling the manufacturing constraints of multi-axis printing, i.e., self-support of structure and overlap/gap-free of continuous fibers. Four numerical examples are presented to demonstrate the effectiveness of the algorithm, especially showing outstanding performances than designs for traditional three-axis 3D printing.}, archive = {J_CMAME}, author = {Kaixian Liang and Jikai Liu and Shuzhi Xu and Yifan Guo}, doi = {10.1016/j.cma.2025.118419}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118419}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {DGTO: Derivable geodesics-coupled topology optimization for multi-axis 3D printing of continuous fiber-reinforced spatial structures}, volume = {448}, year = {2026}, }
• (2026). Active evolutionary gaussian process for structural large-scale full-field reliability analysis and critical domain prognosis with only few initial samples. CMAME, 448, 118418. (www)

  @article{ , abstract = {Reliability analysis is crucial for ensuring structural integrity, yet it requires repeated, time-consuming evaluations of responses and multivariate limit state functions, while struggling to provide full-field estimations efficiently. Therefore, we propose an active evolutionary reduced Gaussian Process framework ( AER-GP ) for fast and large-scale full-field reliability analysis and critical domain prognosis, utilizing only a few initial samples. In which we first define a novel probability indicator of erroneously evaluating the sign of the minimum of the full-field limit state function. Based on this, we then develop an efficient convergence criterion that relies on the expected error in failure probability estimates. Furthermore, we advance the dual order-reduced Gaussian process coupled Monte Carlo methods to accurately predict the large-scale full-field stochastic response under material and load uncertainty. Where the preliminary design of experiments starts from a significantly small number of sets (e.g.,5), and is progressively added by those samples containing the most information to distinguish the failure boundary. More importantly, we propose a novel mechanism for structural critical domain prognosis mechanism based on the failure probability of each single material point within the entire field, and make accurate critical domain prognosis. Real-world examples, including a car wheel hub and a single-tower cable-stayed bridge, demonstrate that the proposed algorithm achieves high prediction accuracy, computational efficiency, and robustness in large-scale structural reliability analysis and critical domain prognosis. It consistently outperforms widely used methods such as AK-MCS, EFF-MCS, ERF-MCS, and H-MCS. Notably, the proposed AER-GP method requires only a small number of initial DoE samples (fewer than 40), and through adaptive learning, it can reliably produce results with very low errors (typically less than 2%).}, archive = {J_CMAME}, author = {Xinlong Li and Chensen Ding}, doi = {10.1016/j.cma.2025.118418}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118418}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Active evolutionary gaussian process for structural large-scale full-field reliability analysis and critical domain prognosis with only few initial samples}, volume = {448}, year = {2026}, }
• (2026). Graph neural network surrogates for contacting deformable bodies with necessary and sufficient contact detection. CMAME, 448, 118413. (www)

  @article{ , abstract = {Surrogate models for the rapid inference of nonlinear boundary value problems in mechanics are helpful in a broad range of engineering applications. However, effective surrogate modeling of applications involving the contact of deformable bodies, especially in the context of varying geometries, is still an open issue. In particular, existing methods are confined to rigid body contact or, at best, contact between rigid and soft objects with well-defined contact planes. Furthermore, they employ contact or collision detection filters that serve as a rapid test but use only the necessary and not sufficient conditions for detection. In this work, we present a graph neural network architecture that utilizes continuous collision detection and, for the first time, incorporates sufficient conditions designed for contact between soft deformable bodies. We test its performance on two benchmarks, including a problem in soft tissue mechanics of predicting the closed state of a bioprosthetic aortic valve. We find a regularizing effect on adding additional contact terms to the loss function, leading to better generalization of the network. These benefits hold for simple contact at similar planes and element normal angles, and complex contact at differing planes and element normal angles. We also demonstrate that the framework can handle varying reference geometries. However, such benefits come with high computational costs during training, resulting in a trade-off that may not always be favorable. We quantify the training cost and the resulting inference speedups on various hardware architectures. Importantly, our graph neural network implementation results in up to a hundred- to thousand-fold speedup on GPU, and twenty- to two hundred-fold speedup on CPU for our benchmark problems at inference.}, archive = {J_CMAME}, author = {Vijay K. Dubey and Collin E. Haese and Osman Gültekin and David Dalton and Manuel K. Rausch and Jan Fuhg}, doi = {10.1016/j.cma.2025.118413}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118413}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Graph neural network surrogates for contacting deformable bodies with necessary and sufficient contact detection}, volume = {448}, year = {2026}, }
• (2026). Quantum neural network-assisted topology optimization: Concept and implementation with parameterized quantum circuits. CMAME, 448, 118411. (www)

  @article{ , abstract = {This study proposes a quantum machine learning-assisted approach to accelerating density-based topology optimization using quantum neural network (QNN), a class of trainable models based on parameterized quantum circuits (PQCs). The proposed framework extracts key features from classical data, starting by encoding the corresponding finite element analysis results, i.e., strain energy, sensitivity, and design variables from early design iterations obtained using the standard optimizer. Then, the PQCs are fine-tuned using minimization of the binary cross-entropy loss function, enabling the model to learn the mapping between input features and optimal design. Specifically, the framework consists of two main stages. First, an offline training stage where the QNN is calibrated using precomputed iterative results from a relatively coarse mesh obtained through the standard optimizer, establishing pattern recognition between input features and the final design variables. The second is an online stage where the trained QNN model is integrated with the standard optimizer to accelerate the final design. Numerical results show that QNN requires only a small number of qubits, and once trained on coarse meshes with several different boundary conditions, can effectively integrate with standard optimizers to predict target designs across various configurations, including different resolutions, volume constraints, and loading conditions. Furthermore, the proposed QNN-assisted framework significantly reduces computational time compared to standard iterative approaches, laying the groundwork for solving large-scale problems with near-term quantum computers.}, archive = {J_CMAME}, author = {Naruethep Sukulthanasorn and Kenjiro Terada}, doi = {10.1016/j.cma.2025.118411}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118411}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Quantum neural network-assisted topology optimization: Concept and implementation with parameterized quantum circuits}, volume = {448}, year = {2026}, }
• (2026). Generative emulation of chaotic dynamics with coherent prior. CMAME, 448, 118410. (www)

  @article{ , abstract = {Data-driven emulation of nonlinear dynamics is challenging due to long-range skill decay that often produces physically unrealistic outputs. Recent advances in generative modeling aim to address these issues by providing uncertainty quantification and correction. However, the quality of generated simulation remains heavily dependent on the choice of conditioning prior. In this work, we present an efficient generative framework for nonlinear dynamics emulation, connecting principles of turbulence with diffusion-based modeling: Cohesion. Our method estimates large-scale coherent structure of the underlying dynamics as guidance during the denoising process, where small-scale fluctuation in the flow is then resolved. These coherent prior are efficiently approximated using reduced-order models, such as deep Koopman operators, that allow for rapid generation of long prior sequences while maintaining stability over extended forecasting horizon. With this gain, we can reframe forecasting as trajectory planning, a common task in reinforcement learning, where conditional denoising is performed once over entire sequences, minimizing the computational cost of autoregressive-based generative methods. Numerical evaluations on chaotic systems of increasing complexity, including Kolmogorov flow, shallow water equations, and subseasonal-to-seasonal climate dynamics, demonstrate Cohesion superior long-range forecasting skill that can efficiently generate physically-consistent simulations, even in the presence of partially-observed guidance.}, archive = {J_CMAME}, author = {Juan Nathaniel and Pierre Gentine}, doi = {10.1016/j.cma.2025.118410}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118410}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Generative emulation of chaotic dynamics with coherent prior}, volume = {448}, year = {2026}, }
• (2026). Concurrent 3D topology optimization method for hierarchical hybrid structures under static and dynamic loads with CPU-GPU heterogeneous parallelism. CMAME, 448, 118408. (www)

  @article{ , abstract = {Topology optimization of 3D hierarchical hybrid structures (HHS) is constrained by the coupling of high-dimensional design spaces and multiscale computational complexity, often addressed by restricting certain designable components, which limits the full exploration of the design space and realization of performance potential. This paper proposes a novel concurrent topology optimization method for 3D-HHS, achieving concurrent optimization of all designable components, including macroscopic topology, substructural topology, and their spatial distribution, under static and dynamic loads. This approach significantly expands the design space, enhancing the mechanical performance of hierarchical structures. To address the computational challenges of large-scale 3D problems, we employ CPU-GPU heterogeneous parallel computing to improve the efficiency of structural response and sensitivity analysis. Numerical examples demonstrate that this method delivers superior 3D-HHS designs with markedly improved optimization efficiency, providing an innovative solution for efficient 3D structural optimization.}, archive = {J_CMAME}, author = {Yunfei Liu and Ruxin Gao and Ying Li and Daining Fang}, doi = {10.1016/j.cma.2025.118408}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118408}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Concurrent 3D topology optimization method for hierarchical hybrid structures under static and dynamic loads with CPU-GPU heterogeneous parallelism}, volume = {448}, year = {2026}, }
• (2026). Configuration-decoupled concurrent topology optimization of heterogeneous lattice structures. CMAME, 448, 118405. (www)

  @article{ , abstract = {Multi-configuration lattice structures have recently been introduced into structural optimization due to their broadly tunable physical properties. Traditional methods for multi-configuration lattice optimization employ extreme strategies of complete fusion or separation, leading to a trade-off between optimality and scalability that has not been fully addressed in the existing literature. The paper suggests decomposing the lattice library into pairs of lattices, through which multi-configuration lattice optimization is decoupled into the concurrent optimization of iso-value, combination category, and ratio within combination. A novel hybrid interpolation scheme is proposed to describe the effective mechanical behavior of the configuration-decoupled lattices. In this approach, polynomial models are employed to characterize the performance of individual lattice combinations, while the Uniform Multiphase Materials Interpolation model is used to integrate the contributions of all combinations. Benchmark experiments, including full-scale simulations, are conducted to validate the effectiveness of the framework. The proposed method enables rapid convergence to configuration layouts that align with the principal stress orientations. Compared to single- and dual-configuration designs, it achieves compliance reductions of 61.0 % and 26.2 %, respectively, approaching the performance of density-based topology optimization. Extended numerical experiments reveal the joint influence of resolution and configuration count on the overall performance. This method achieves a better trade-off between optimality and extensibility, enabling more flexible utilization of large lattice databases in practical engineering fields.}, archive = {J_CMAME}, author = {Xinze Shen and Changdong Zhang and Wenhe Liao and Dawei Li and Tingting Liu}, doi = {10.1016/j.cma.2025.118405}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118405}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Configuration-decoupled concurrent topology optimization of heterogeneous lattice structures}, volume = {448}, year = {2026}, }
• (2026). Intrepid MCMC: Metropolis-hastings with exploration. CMAME, 448, 118402. (www)

  @article{ , abstract = {In engineering examples, one often encounters the need to sample from unnormalized distributions with complex shapes that may also be implicitly defined through a physical or numerical simulation model, making it computationally expensive to evaluate the associated density function. For such cases, MCMC has proven to be an invaluable tool. Random-walk Metropolis Methods (also known as Metropolis-Hastings (MH)), in particular, are highly popular for their simplicity, flexibility, and ease of implementation. However, most MH algorithms suffer from significant limitations when attempting to sample from distributions with multiple modes (particularly disconnected ones). In this paper, we present Intrepid MCMC - a novel MH scheme that utilizes a simple coordinate transformation to significantly improve the mode-finding ability and convergence rate to the target distribution of random-walk Markov chains while retaining most of the simplicity of the vanilla MH paradigm. Through multiple examples, we showcase the improvement in the performance of Intrepid MCMC over vanilla MH for a wide variety of target distribution shapes. We also provide an analysis of the mixing behavior of the Intrepid Markov chain, as well as the efficiency of our algorithm for increasing dimensions. A thorough discussion is presented on the practical implementation of the Intrepid MCMC algorithm. Finally, its utility is highlighted through a Bayesian parameter inference problem for a two-degree-of-freedom oscillator under free vibration.}, archive = {J_CMAME}, author = {Promit Chakroborty and Michael D. Shields}, doi = {10.1016/j.cma.2025.118402}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118402}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Intrepid MCMC: Metropolis-hastings with exploration}, volume = {448}, year = {2026}, }
• (2026). Two-level overlapping additive schwarz preconditioner for training scientific machine learning applications. CMAME, 448, 118400. (www)

  @article{ , abstract = {We introduce a novel two-level overlapping additive Schwarz preconditioner for accelerating the training of scientific machine learning applications. The design of the proposed preconditioner is motivated by the nonlinear two-level overlapping additive Schwarz preconditioner. The neural network parameters are decomposed into groups (subdomains) with overlapping regions. In addition, the network’s feed-forward structure is indirectly imposed through a novel subdomain-wise synchronization strategy and a coarse-level training step. Through a series of numerical experiments, which consider physics-informed neural networks and operator learning approaches, we demonstrate that the proposed two-level preconditioner significantly speeds up the convergence of the standard (LBFGS) optimizer while also yielding more accurate machine learning models. Moreover, the devised preconditioner is designed to take advantage of model-parallel computations, which can further reduce the training time.}, archive = {J_CMAME}, author = {Youngkyu Lee and Alena Kopaničáková and George Em Karniadakis}, doi = {10.1016/j.cma.2025.118400}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118400}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Two-level overlapping additive schwarz preconditioner for training scientific machine learning applications}, volume = {448}, year = {2026}, }
• (2026). Overlapping schwarz preconditioners for isogeometric discretizations of acoustic wave problems. CMAME, 448, 118397. (www)

  @article{ , abstract = {The aim of this work is to construct and analyze two-level overlapping additive Schwarz (OAS) preconditioners for isogeometric discretizations of the acoustic wave equation with absorbing boundary conditions. Both Collocation and Galerkin isogeometric methods are employed for space discretization, while time advancing is performed by means of a Newmark implicit scheme. The linear systems to be solved at each time step are ill conditioned, especially in case of highly regular splines, thus their solution requires the use of effective preconditioners. Two-level OAS solvers consist of partitioning the domain into overlapping subdomains, solving independent local problems on each subdomain and an additional coarse problem associated with the subdomain mesh. Several two-dimensional numerical results validate our theoretical estimates, showing the scalability and quasi-optimality of the algorithms proposed. We also investigate numerically the robustness of the OAS preconditioners with respect to the spline polynomial degree, the spline regularity and the overlap parameter.}, archive = {J_CMAME}, author = {Elena Zampieri and Simone Scacchi and Luca F. Pavarino}, doi = {10.1016/j.cma.2025.118397}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118397}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Overlapping schwarz preconditioners for isogeometric discretizations of acoustic wave problems}, volume = {448}, year = {2026}, }
• (2026). Latent space modeling of parametric and time-dependent PDEs using neural ODEs. CMAME, 448, 118394. (www)

  @article{ , abstract = {Partial Differential Equations (PDEs) are central to science and engineering. Since solving them is computationally expensive, a lot of effort has been put into approximating their solution operator via both traditional and recently increasingly Deep Learning (DL) techniques. In this paper, we propose an autoregressive and data-driven method using the analogy with classical numerical solvers for time-dependent, parametric and (typically) nonlinear PDEs. We present how Dimensionality Reduction (DR) can be coupled with Neural Ordinary Differential Equations (NODEs) in order to learn the solution operator of arbitrary PDEs accounting both for (continuous) time and parameter dependency. The idea of our work is that it is possible to map the high-fidelity (i.e., high-dimensional) PDE solution space into a reduced (low-dimensional) space, which subsequently exhibits dynamics governed by a (latent) Ordinary Differential Equation (ODE). Solving this (easier) ODE in the reduced space allows avoiding solving the PDE in the high-dimensional solution space, thus decreasing the computational burden for repeated calculations for e.g., uncertainty quantification or design optimization purposes. The main outcome of this work is the importance of exploiting DR as opposed to the recent trend of building large and complex architectures: we show that by leveraging DR we can deliver not only more accurate predictions, but also a considerably lighter and faster DL model compared to existing methodologies.}, archive = {J_CMAME}, author = {Alessandro Longhi and Danny Lathouwers and Zoltán Perkó}, doi = {10.1016/j.cma.2025.118394}, journal = {Computer Methods in Applied Mechanics and Engineering}, month = {1}, pages = {118394}, shortjournal = {Comput. Meth. Appl. Mech. Eng.}, title = {Latent space modeling of parametric and time-dependent PDEs using neural ODEs}, volume = {448}, year = {2026}, }

COMCOM - 5

• (2025). Revisiting the problem of optimizing spreading factor allocations in LoRaWAN: From theory to practice. COMCOM, 243, 108321. (www)

  @article{ , abstract = {This paper revisits the problem of optimizing LoRa network success probability by proposing an optimized allocation strategy for Spreading Factors (SFs) under both uniform and Gaussian network deployments with a single or multiple gateways. More specifically, we solve the problem of finding the best SF allocations in dense network deployments whose EDs are first assigned with the minimum SF. Theoretical models are developed to quantify the success probability of transmissions, considering the capture effect as well as intra- and inter-SF interference. A mathematical optimization framework is introduced to determine the optimal SF distribution that maximizes the average probability of packet reception. The problem is solved using Mixed Integer Linear Programming (MILP), and then evaluated using simulations. Even though optimal SF allocation strategies have been proposed in the literature, no practical insights have been discovered and no real-world deployments have been considered. To this extent, the practical benefits of using improved or optimal SF settings are discovered in this paper. Simulation results confirm the theoretical findings while they demonstrate an up to 10 percentage points improvements in Packet Reception Ratio (PRR) in the real-world use-case.}, archive = {J_COMCOM}, author = {Dimitrios Zorbas and Aruzhan Sabyrbek and Luigi Di Puglia Pugliese}, doi = {10.1016/j.comcom.2025.108321}, journal = {Computer Communications}, month = {11}, pages = {108321}, shortjournal = {Comput. Commun.}, title = {Revisiting the problem of optimizing spreading factor allocations in LoRaWAN: From theory to practice}, volume = {243}, year = {2025}, }
• (2025). Energy consumption optimization in UAV-assisted multi-layer mobile edge computing with active transmissive RIS. COMCOM, 243, 108320. (www)

  @article{ , abstract = {Unmanned Aerial Vehicle (UAV)-assisted edge computing provides low-latency and low-energy consumption computing capabilities for sparsely distributed Internet of Things (IoT) networks. In addition, the assisted UAVs provide line-of-sight links to further improve communication quality. However, the existing offloading strategies have low efficiency and high costs. Motivated by this, we propose a novel UAV-assisted multi-layer mobile edge computing network with active transmissive reconfigurable intelligent surface (RIS). The introduced an active transmissive RIS not only receives data from UAVs but also performs computing functionality. We establish an optimization to minimize the total system energy consumption under delay constraints by jointly planning UAV positions and allocating computing bits, sub-carriers, time slots, transmission power, and RIS transmission coefficient. To tackle this problem, we first use the block coordinate descent (BCD) algorithm to decouple it into four sub-problems. Then, we solve them by adopting successive convex approximation (SCA), difference-convex (DC) programming, and introducing slack variables. Experimental results demonstrate that the proposed network is superior to the other five baselines concerning energy consumption reduction. Also, the influences of system parameters are verified, including the number of IoT devices, the number of RIS elements, and the delay threshold.}, archive = {J_COMCOM}, author = {Kexin Yang and Yaxi Liu and Boxin He and Jiahao Huo and Wei Huangfu}, doi = {10.1016/j.comcom.2025.108320}, journal = {Computer Communications}, month = {11}, pages = {108320}, shortjournal = {Comput. Commun.}, title = {Energy consumption optimization in UAV-assisted multi-layer mobile edge computing with active transmissive RIS}, volume = {243}, year = {2025}, }
• (2025). A preemptive task unloading scheme based on second optional unloading in cloud-fog collaborative networks. COMCOM, 243, 108315. (www)

  @article{ , abstract = {Long-distance data transmission between Internet of Things (IoT) devices and remote cloud center often leads to unacceptable latency for certain tasks. Fog computing has emerged as a promising solution for low-latency tasks. Consequently, the concept of cloud-fog collaborative networks has garnered significant attention. However, existing research primarily focuses on heterogeneous tasks, overlooking the crucial aspect of considering both task priority and second unloading. To address this gap, this paper proposes a novel task unloading scheme that concurrently takes preemptive priority and second optional unloading into account. In this scheme, delay-sensitive tasks (DSTs) are given preemptive priority over delay-tolerant tasks (DTTs). Furthermore, some DTTs may undergo preprocessing in the fog layer to optimize resource utilization. Moreover, tasks encountering blocking or preemption in the fog layer can also be secondarily unloaded to the cloud layer. In this framework, we devise a four-dimensional Markov chain (4DMC) to model and analyze this process. Through numerical experiments, we assess performance indicators under various parameters. Ultimately, our proposed strategy is compared with the unloading scheme that does not incorporate second unloading through both numerical analysis and simulation validation. The results indicate that our scheme notably enhances the throughput of DTTs, albeit at a marginal performance trade-off.}, archive = {J_COMCOM}, author = {Yuan Zhao and Hongmin Gao and Shuaihua Liu}, doi = {10.1016/j.comcom.2025.108315}, journal = {Computer Communications}, month = {11}, pages = {108315}, shortjournal = {Comput. Commun.}, title = {A preemptive task unloading scheme based on second optional unloading in cloud-fog collaborative networks}, volume = {243}, year = {2025}, }
• (2025). ScaleIP: A hybrid autoscaling of VoIP services based on deep reinforcement learning. COMCOM, 243, 108314. (www)

  @article{ , abstract = {Adaptive resource provisioning has become crucial for cloud-based applications, especially those managing real-time traffic like Voice over IP (VoIP), which experience rapidly fluctuating workloads. Traditional static provisioning methods often fall short in these dynamic environments, leading to inefficiencies and potential service disruptions. Existing solutions struggle to maintain performance under varying traffic conditions, particularly for time-sensitive applications. This paper introduces ScaleIP, a hybrid autoscaling solution for containerized VoIP services that offers real-time adaptability and efficient resource management. ScaleIP leverages Deep Reinforcement Learning to make dynamic and efficient scaling decisions, improving call latency, increasing the number of successfully routed calls, and maximizing resource utilization. We evaluated ScaleIP through extensive experiments conducted on a real testbed utilizing the customer Call Detail Record (CDR) from 2023 provided by World Direct, encompassing over 89 million calls. The results show that ScaleIP consistently maintains call latency below 2 s, increases the number of successfully routed calls by 3.26 ×, and increases the resource utilization up to 60 % compared to state-of-the-art autoscaling methods.}, archive = {J_COMCOM}, author = {Zahra Najafabadi Samani and Juan Aznar Poveda and Dominik Gratz and Rene Hueber and Philipp Kalb and Thomas Fahringer}, doi = {10.1016/j.comcom.2025.108314}, journal = {Computer Communications}, month = {11}, pages = {108314}, shortjournal = {Comput. Commun.}, title = {ScaleIP: A hybrid autoscaling of VoIP services based on deep reinforcement learning}, volume = {243}, year = {2025}, }
• (2025). Beyond performance comparing the costs of applying deep and shallow learning. COMCOM, 243, 108312. (www)

  @article{ , abstract = {The rapid growth of mobile network traffic and the emergence of complex applications, such as self-driving cars and augmented reality, demand ultra-low latency, high throughput, and massive device connectivity, which traditional network design approaches struggle to meet. These issues were initially addressed in Fifth-Generation (5G) and Beyond-5G (B5G) networks, where Artificial Intelligence (AI), particularly Deep Learning (DL), is proposed to optimize the network and to meet these demanding requirements. However, the resource constraints and time limitations inherent in telecommunication networks raise questions about the practicality of deploying large Deep Neural Networks (DNNs) in these contexts. This paper analyzes the costs of implementing DNNs by comparing them with shallow ML models across multiple datasets and evaluating factors such as execution time and model interpretability. Our findings demonstrate that shallow ML models offer comparable performance to DNNs, with significantly reduced training and inference times, achieving up to 90% acceleration. Moreover, shallow models are more interpretable, as explainability metrics struggle to agree on feature importance values even for high-performing DNNs.}, archive = {J_COMCOM}, author = {Rafael Teixeira and Leonardo Almeida and Pedro Rodrigues and Mário Antunes and Diogo Gomes and Rui L. Aguiar}, doi = {10.1016/j.comcom.2025.108312}, journal = {Computer Communications}, month = {11}, pages = {108312}, shortjournal = {Comput. Commun.}, title = {Beyond performance comparing the costs of applying deep and shallow learning}, volume = {243}, year = {2025}, }

COR - 25

• (2026). A novel mathematical model for the scheduling of a zero inventory production: An application of process scheduling in fog computing. COR, 185, 107284. (www)

  @article{ , abstract = {One of the main production-related costs in manufacturing is inventory cost since manufacturing firms allocate a vast area to raw material, semi-processed, and final products in production lines and warehouses. Reducing the volume of these inventories leads to lower production-related costs. This paper presents a novel mathematical model for zero-inventory production scheduling. In this model, the jobs arrive at fixed times and are scheduled on a set of unrelated machines. The jobs have different operations that need to be processed one by one. Since the system has zero inventory, the jobs must be processed immediately upon arrival. Also, whenever a job’s operation is complete, the following operation must instantly start (no wait time). That operation is outsourced if no machines are available to process any of the job’s operations. The jobs’ operations are dispatched to the machines from a dispatching center, and there is a latency between the dispatching center, the machines, and the outsourcing center. We present a mixed-integer non-linear programming (MINLP) model to formulate this problem. Then, the MINLP model is turned into a mixed-integer linear programming (MILP) model by linearizing its constraints. Since many production scheduling problems are known to be NP-hard, particularly those involving unrelated parallel machines, precedence constraints, and time-dependent decisions like ours, we adopt two metaheuristics to solve the problem for large-scale cases where exact methods are computationally inefficient. The first is a Genetic Algorithm (GA), and the second is a Teaching-Learning-Based Optimization (TLBO) algorithm. The performance of these algorithms is tested against the optimal solutions obtained from CPLEX for a set of small-scale problems. We consider a real case study, an image processing system, to validate the proposed developments (the MILP model and the GA). The results show that the presented model and algorithm can reduce the system’s total cost by about 12.57% compared to the existing online dispatching rules.}, archive = {J_COR}, author = {Mani Sharifi and Sharareh Taghipour and Abdolreza Abhari and Maciej Rysz}, doi = {10.1016/j.cor.2025.107284}, journal = {Computers & Operations Research}, month = {1}, pages = {107284}, shortjournal = {Comput. Oper. Res.}, title = {A novel mathematical model for the scheduling of a zero inventory production: An application of process scheduling in fog computing}, volume = {185}, year = {2026}, }
• (2026). A unified approach to extract interpretable rules from tree ensembles via integer programming. COR, 185, 107283. (www)

  @article{ , abstract = {Tree ensembles are widely used machine learning models, known for their effectiveness in supervised classification and regression tasks. Their performance derives from aggregating predictions of multiple decision trees, which are renowned for their interpretability properties. However, tree ensemble models do not reliably exhibit interpretable output. Our work aims to extract an optimized list of rules from a trained tree ensemble, providing the user with a condensed, interpretable model that retains most of the predictive power of the full model. Our approach consists of solving a set partitioning problem formulated through Integer Programming. The extracted list of rules is unweighted and defines a partition of the training data, assigning each instance to exactly one rule, and thereby simplifying the explanation process. The proposed method works with tabular or time series data, for both classification and regression tasks, and its flexible formulation can include any arbitrary loss or regularization functions. Our computational experiments offer statistically significant evidence that our method performs comparably to several rule extraction methods in terms of predictive performance and fidelity towards the tree ensemble. Moreover, we empirically show that the proposed method effectively extracts interpretable rules from tree ensembles that are designed for time series data.}, archive = {J_COR}, author = {Lorenzo Bonasera and Emilio Carrizosa}, doi = {10.1016/j.cor.2025.107283}, journal = {Computers & Operations Research}, month = {1}, pages = {107283}, shortjournal = {Comput. Oper. Res.}, title = {A unified approach to extract interpretable rules from tree ensembles via integer programming}, volume = {185}, year = {2026}, }
• (2026). Optimizing high-tech product take-back schemes in a closed-loop supply chain. COR, 185, 107282. (www)

  @article{ , abstract = {Frequent product development is a solution to the shortened product lifecycles in the consumer electronics industry. It enables companies to maintain competitiveness and strengthen their market share. However, environmental concerns bring reverse logistics practices into focus. A take-back policy is a strategic reverse logistics activity known to foster market share; however, it poses various challenges and uncertainties. Considering uncertain demand, we introduced an innovative adoption model with two distinct take-back policies, trade-in and credit, to address challenges in multi-generation production planning. Inspired by real-world practices of companies like Apple and Samsung, our model first examines how trade-in programs drive repeat purchases and enhance market share, with credit-based programs to attract new customers. It then captures changes in demand, production planning, recovery decisions, and internal competition among multiple product generations. Distinct from previous conclusions, this study explores how producers can strategically manage demand for new generations to slow diffusion, thereby increasing refurbishment and recycling volumes over time. Our findings highlight the pivotal role of adaptive pricing strategies and production scalability in maximizing profitability and promoting sustainability in competitive high-tech industries.}, archive = {J_COR}, author = {Fatemeh Keshavarz-Ghorbani and Mohamad Y. Jaber and Seyed Hamid Reza Pasandideh}, doi = {10.1016/j.cor.2025.107282}, journal = {Computers & Operations Research}, month = {1}, pages = {107282}, shortjournal = {Comput. Oper. Res.}, title = {Optimizing high-tech product take-back schemes in a closed-loop supply chain}, volume = {185}, year = {2026}, }
• (2026). Q-learning-based hyper-heuristic algorithm for open dimension irregular packing problems. COR, 185, 107279. (www)

  @article{ , abstract = {Heuristic methods provide a computationally efficient framework for addressing two-dimensional irregular packing problems, particularly in resource-constrained industrial settings. As a typical combinatorial optimization problem, irregular packing exhibits exponential growth in computational complexity with increasing workpiece counts, while the solution space dynamically reconfigures due to geometric variability among workpieces. Although heuristic algorithms can generate feasible layouts within acceptable timeframes, their reliance on fixed search rule limits adaptability to diverse scenarios, necessitating more flexible approaches. In this paper, a hyper-heuristic algorithm based on Q-Learning is proposed to solve open dimension packing problems, including one-open and two-open dimension problems. Q-Learning is adopted as the high-level strategy for its ability to optimize low-level heuristic selection through reward-driven experience accumulation. The method incorporates a mixed encoding method for solution representation, four specialized low-level heuristic operators, a linear population decline mechanism, and an elite preservation strategy to balance exploration–exploitation. The Q-Learning controller dynamically selects operators by updating the Q-table based on Bellman’s equation. The proposed algorithm is compared to some advanced algorithms in general datasets. The results show that our method has better performance and applicability.}, archive = {J_COR}, author = {Yongchun Wang and Qingjin Peng and Zhen Wang and Shuiquan Huang and Zhengkai Xu and Chuanzhen Huang and Baosu Guo}, doi = {10.1016/j.cor.2025.107279}, journal = {Computers & Operations Research}, month = {1}, pages = {107279}, shortjournal = {Comput. Oper. Res.}, title = {Q-learning-based hyper-heuristic algorithm for open dimension irregular packing problems}, volume = {185}, year = {2026}, }
• (2026). Hybrid modelling using simulation and machine learning in healthcare. COR, 185, 107278. (www)

  @article{ , abstract = {Modelling & Simulation (M&S) and Machine Learning (ML) methodologies have undergone significant advancements, enabling transformative applications across various industries. The integration of M&S and ML into a Hybrid M&S-ML approach leverages the unique strengths of both fields, offering enhanced model precision, improved efficiency, and more effective decision support. This review explores the increasing convergence of ML algorithms with traditional M&S methods- namely Agent-Based Modelling & Simulation, Discrete Event Simulation, and System Dynamics- in healthcare applications. Through a systematic review of 90 relevant studies, this article provides a comprehensive synthesis of the current state-of-the-art Hybrid M&S-ML in healthcare. Specifically, it examines the M&S and ML methodologies employed, associated software tools and programming languages, analyses integration patterns and data exchange mechanisms, and explores application domains, as well as the types and motivations for hybridisation. Key findings highlight prominent methodological and technical trends, as well as opportunities for combining M&S with ML to address healthcare challenges. These insights provide direction for modellers and data scientists in developing hybrid M&S–ML approaches that more effectively combine simulation capabilities with data-driven learning. The review also demonstrates the potential of such approaches to advance methodological innovation and support evidence-based decision-making in diverse healthcare contexts.}, archive = {J_COR}, author = {Ali Ahmadi and Masoud Fakhimi and Carin Magnusson}, doi = {10.1016/j.cor.2025.107278}, journal = {Computers & Operations Research}, month = {1}, pages = {107278}, shortjournal = {Comput. Oper. Res.}, title = {Hybrid modelling using simulation and machine learning in healthcare}, volume = {185}, year = {2026}, }
• (2026). A note on battery swapping policies in the electric vehicle routing problem with time windows and battery swapping vehicles. COR, 185, 107277. (www)

  @article{ , abstract = {Çatay and Sadati [An improved matheuristic for solving the electric vehicle routing problem with time windows and synchronized mobile charging/battery swapping. Computers & Operations Research 159, 106310, 2023] explores a variant of the Electric Vehicle Routing Problem with Time Windows that incorporates mobile chargers for recharging electric vehicles (EVs) at selected locations while serving customers. The authors propose a matheuristic method to address this problem and its special case, where EV batteries are swapped in constant time instead of being recharged over variable durations. While comparing their results with those in the literature, the authors overlook a critical assumption regarding the swapping policy, potentially causing confusion in interpreting the findings. This note addresses the issue, clarifies the overlooked assumption, and updates the results that do not align with the actual scenario in the literature. Furthermore, it introduces two new battery swapping policies and presents an extensive computational study to offer new insights on synchronized mobile battery swapping.}, archive = {J_COR}, author = {Bülent Çatay and İhsan Sadati}, doi = {10.1016/j.cor.2025.107277}, journal = {Computers & Operations Research}, month = {1}, pages = {107277}, shortjournal = {Comput. Oper. Res.}, title = {A note on battery swapping policies in the electric vehicle routing problem with time windows and battery swapping vehicles}, volume = {185}, year = {2026}, }
• (2026). An adaptive K-means and reinforcement learning (RL) algorithm to effective vaccine distribution. COR, 185, 107275. (www)

  @article{ , abstract = {We present a new adaptive reinforcement learning (RL) approach, integrated with a K-means clustering algorithm and guided by simulated annealing, to address the capacitated vehicle routing for vaccine distribution (CVRVD) problem. This integrated method provides an efficient and scalable solution for optimizing vaccine distribution logistics. By incorporating cost factors related to travel distance, inventory levels, and penalty terms – while adhering to delivery time windows – our approach improves both operational efficiency and vaccine allocation effectiveness. Experimental results demonstrate that our K-means supported RL algorithm significantly outperforms traditional solvers in tackling this NP-hard problem, particularly in large-scale scenarios. Specifically, our approach can efficiently solve CVRVD instances with up to 1,000 facilities—scenarios that are computationally intractable for exact methods. We demonstrate the effectiveness of the adaptive K-means supported RL algorithm using data from New Jersey, USA, where facility-level vaccination data were available through the state’s Immunization Information System. Beyond vaccine distribution, our method has broad applicability in logistics and transportation, enabling more efficient and cost-effective allocation of critical resources such as vaccines and medical supplies.}, archive = {J_COR}, author = {Elson Cibaku and İ. Esra Büyüktahtakın}, doi = {10.1016/j.cor.2025.107275}, journal = {Computers & Operations Research}, month = {1}, pages = {107275}, shortjournal = {Comput. Oper. Res.}, title = {An adaptive K-means and reinforcement learning (RL) algorithm to effective vaccine distribution}, volume = {185}, year = {2026}, }
• (2026). A Q-learning-based evolutionary algorithm for solving the low-carbon multi-objective flexible job shop scheduling problem. COR, 185, 107266. (www)

  @article{ , abstract = {In recent years, how to reduce energy consumption at the manufacturing system level in the low-carbon multi-objective flexible job shop scheduling problem (LCM-FJSP) has received significant attention. In this research, a model with the maximum completion time, total machine workload and total carbon emissions is built. Moreover, a Q-learning-based adaptive weight-adjusted decomposition evolutionary algorithm (QMOEA/D-AWA) is proposed. In the QMOEA/D-AWA, an initialization strategy with four heuristic initial rules for obtaining high-quality population, a variable neighborhood search strategy with four problem-specific local search methods for enhancing exploration and a Q-learning-based parameter adaptive strategy for automatically determining the number of neighborhood solutions are designed. To validate the effectiveness of the proposed QMOEA/D-AWA, it is compared with five state-of-the-art algorithms on 15 instances. In the statistical analysis, the QMOEA/D-AWA obtains the overwhelming metric results in 10 instances. In the visual analysis, the completion time is reduced by 3.74%, the total workload is reduced by 3.94%, and the carbon emissions are reduced by 5.94%.}, archive = {J_COR}, author = {Zhixue Wang and Maowei He and Hanning Chen and Yabao Hu and Yelin Xia}, doi = {10.1016/j.cor.2025.107266}, journal = {Computers & Operations Research}, month = {1}, pages = {107266}, shortjournal = {Comput. Oper. Res.}, title = {A Q-learning-based evolutionary algorithm for solving the low-carbon multi-objective flexible job shop scheduling problem}, volume = {185}, year = {2026}, }
• (2026). A benders-branch-and-cut methodology for global cargo vessel traffic prediction given declining arctic sea ice and changing risks. COR, 185, 107265. (www)

  @article{ , abstract = {Global warming has led to declining sea-ice in the Arctic Ocean, making it easier for ice-class vessels to navigate Arctic waters for greater portions of the year. As sailing conditions in these waters improve over coming decades, these passageways are expected to open for larger portions of the year and to become increasingly viable options for unsupported transit and even open-water vessels. This paper proposes a Benders-branch-and-cut methodology for estimating changes in global maritime cargo flow patterns under future climate scenarios with declining Arctic sea ice. The model accounts for changing incident risk along Arctic passageways and corresponding ice-class vessel and icebreaker escort requirements, lower speeds, increased insurance premiums, higher accident probabilities, and constraints on path-based maximum risk exposure. The resulting mixed-integer program involves path-based, continuous decision variables. The solution technique is applied on a model of the global maritime container network including 80 ports, 76 routes, 426 links and 4,303 legs associated with the world’s largest carrier alliance. Embedded acceleration techniques and a label-correcting algorithm that employs specialized fathoming rules for a non-additive, constrained path subproblem enable solution at this global scale. The outcome is an estimate of seasonal future global maritime trade flows along key global routes and through ports predicted under six climate-related scenarios. Results illustrate that the developed model can provide support to companies, nations and regions as they prepare for a changing global landscape and climate.}, archive = {J_COR}, author = {Wenjie Li and Elise Miller-Hooks}, doi = {10.1016/j.cor.2025.107265}, journal = {Computers & Operations Research}, month = {1}, pages = {107265}, shortjournal = {Comput. Oper. Res.}, title = {A benders-branch-and-cut methodology for global cargo vessel traffic prediction given declining arctic sea ice and changing risks}, volume = {185}, year = {2026}, }
• (2026). A deep reinforcement learning approach for dynamic job-shop scheduling problem considering time variable and new job arrivals. COR, 185, 107263. (www)

  @article{ , abstract = {In recent years, the complexity of the production process due to increased demand for customization has greatly increased the difficulty of dynamic job-shop scheduling problem (DJSP). This paper proposes a deep reinforcement learning (DRL) approach to tackle the DJSP based on proximal policy optimization (PPO) algorithm. A novel state representation method that expresses state features as multi-channel images is proposed to simplify the state characterization process. Various heuristic-based priority dispatching rules (PDRs)are used to construct action space. By converting scheduling instances into images and leveraging the spatial pyramid pooling fast (SPPF) module for feature extraction, this model can handle scheduling instances of varying scales and map size-independent processing information matrix to fixed action space. Additionally, a dense reward based on a predefined scheduling region is developed to offer detailed guidance to the agent, enabling more precise and comprehensive policy assessment. Static tests are conducted on well-known benchmarks, and the experimental results indicate that our scheduling model surpasses the performance of the three latest DRL approaches on average. Compared with PDR methods, dynamic experiments demonstrate that the proposed DRL model excels in adaptability and robustness when new tasks arrive and the processing time fluctuates with uncertainty.}, archive = {J_COR}, author = {Haoyang Yu and Wenbin Gu and Na Tang and Zhenyang Guo}, doi = {10.1016/j.cor.2025.107263}, journal = {Computers & Operations Research}, month = {1}, pages = {107263}, shortjournal = {Comput. Oper. Res.}, title = {A deep reinforcement learning approach for dynamic job-shop scheduling problem considering time variable and new job arrivals}, volume = {185}, year = {2026}, }
• (2026). An efficient resource utilization and scheduling strategy for in-service aircraft maintenance and operations. COR, 185, 107262. (www)

  @article{ , abstract = {Optimal scheduling of maintenance activities requires the solution of combinatorial optimization problems that need to be efficiently modeled and solved with optimization techniques. Maintenance scheduling and operations-associated problems in the aviation industry can efficiently enhance competitiveness. In the maintenance and scheduling problem, aircrafts need to undergo tasks for both line (A check) and base (C check) maintenance at various hangers at MRO (Maintenance, Repair and Operations) based on resource availability (both human and material). The determination of the optimal maintenance plan, in terms of allocating the resources to the aircraft, and resource movement from one aircraft to another based on availability and licensed skills in the presence of multiple tasks and capacity constraints so as to obtain maximum utilization of resources at maintenance site and minimize the turnaround time is a complex combinatorial optimization problem. To the best of our knowledge, this work is the first CP (Constraint Programming) based mathematical solution that jointly integrates zone, task precedence, technician-pool sharing, and multi-shift continuity for large-scale aircraft maintenance scheduling. In this article, we proposed an efficient optimization strategy that overcomes many of the drawbacks of the formulation/strategies available in literature and helps in determining efficient execution of maintenance work packages. The proposed strategy is generic, encompassing multi-aircraft, multi-skill and multi-shift scheduling capabilities and is validated on two real scenario business case studies, one each for line maintenance (A check) tasks and base maintenance (C check) tasks, as well as six large-scale synthetic scenarios with up to 20,000 tasks, demonstrating feasibility and scalable performance. The proposed strategy is demonstrated on MRO scheduling and it shows an improvement of up to 30.68% in the turn-around time by incorporating the proposed optimization strategy.}, archive = {J_COR}, author = {Sandeep Singh Chauhan and Likhith Maadhav and Abhijit Dake and Gauthier Brillaud}, doi = {10.1016/j.cor.2025.107262}, journal = {Computers & Operations Research}, month = {1}, pages = {107262}, shortjournal = {Comput. Oper. Res.}, title = {An efficient resource utilization and scheduling strategy for in-service aircraft maintenance and operations}, volume = {185}, year = {2026}, }
• (2026). A data-driven optimization approach for the integrated train scheduling and maintenance planning in high-speed railways. COR, 185, 107261. (www)

  @article{ , abstract = {In railway systems, preventive maintenance plans are essential for ensuring the safety of train operations. However, these tasks are often subject to various disturbances (e.g., bad weather), leading to unpredictable deviations between planned and actual maintenance durations, which can further disrupt train schedules. Unlike most studies that assume constant maintenance durations, this paper introduces a data-driven, two-stage distributionally robust optimization (DRO) model for jointly optimizing train scheduling and maintenance planning. In the first stage, we determine the initial train schedule and maintenance plan. In the second stage, we allow for slight adjustments to train departure and arrival times at each station to accommodate disturbances affecting maintenance tasks. Our objective is to minimize both the expected travel time of trains and the deviation from the planned schedule under worst-case scenarios for maintenance disturbances. To capture the uncertainty of maintenance disturbances, we construct an ambiguity set using historical data and the Wasserstein metric. We show that the proposed two-stage DRO model, formulated over the Wasserstein ambiguity set, can be reformulated into an efficiently solvable equivalent form. Finally, we apply our model to a real-world case study of the Beijing–Guangzhou high-speed railway and compare it with traditional stochastic programming methods, including sample average approximation and robust optimization. The results highlight the efficiency of our approach and provide valuable insights for railway management.}, archive = {J_COR}, author = {Hangyu Ji and Chuntian Zhang and Jiateng Yin and Lixing Yang}, doi = {10.1016/j.cor.2025.107261}, journal = {Computers & Operations Research}, month = {1}, pages = {107261}, shortjournal = {Comput. Oper. Res.}, title = {A data-driven optimization approach for the integrated train scheduling and maintenance planning in high-speed railways}, volume = {185}, year = {2026}, }
• (2026). ROBIST: Robust optimization by iterative scenario sampling and statistical testing. COR, 185, 107260. (www)

  @article{ , abstract = {In this paper, we propose ROBIST , a simple, yet effective, data-driven algorithm for optimization under parametric uncertainty. The algorithm first generates solutions in an iterative manner by sampling and optimizing over a relatively small set of scenarios. Then, using statistical testing, the robustness of the solutions is evaluated, which can be done with a much larger set of scenarios. ROBIST offers a number of practical advantages over existing methods as it is: (i) easy to implement, (ii) able to deal with a wide range of problems and (iii) capable of providing sharp probability guarantees that are easily computable and independent of the dimensions of the problem. Numerical experiments demonstrate the effectiveness of ROBIST in comparison to alternative methods.}, archive = {J_COR}, author = {Justin Starreveld and Guanyu Jin and Dick den Hertog and Roger J.A. Laeven}, doi = {10.1016/j.cor.2025.107260}, journal = {Computers & Operations Research}, month = {1}, pages = {107260}, shortjournal = {Comput. Oper. Res.}, title = {ROBIST: Robust optimization by iterative scenario sampling and statistical testing}, volume = {185}, year = {2026}, }
• (2026). A branch-and-price algorithm for energy aware task scheduling of constellations of nanosatellites. COR, 185, 107259. (www)

  @article{ , abstract = {This paper presents a branch-and-price algorithm for solving the Optimal Network Task Scheduling (ONTS) problem in satellite constellations. The algorithm efficiently manages both constellation tasks that can be performed by any satellite and satellite-specific tasks that must be executed by designated satellites, while considering critical energy constraints. We formulate the problem as a Mixed-Integer Linear Programming (MILP) model and develop a Dantzig–Wolfe decomposition that handles battery management constraints for the satellites at the master level, while addressing constellation-wide coordination requirements in the subproblems. A novel dynamic programming algorithm is proposed to solve the pricing subproblem for constellation tasks, augmented with dual stabilization techniques to improve convergence. Comprehensive computational experiments on realistic instances derived from nanosatellite operations demonstrate the effectiveness of the algorithm. Results show that our structured formulation significantly outperforms a naive approach, particularly for large instances, while effectively balancing workload distribution and energy management across the constellation. This work provides a practical framework for optimizing task scheduling in modern satellite constellations, with direct applications in Earth observation, telecommunications, and scientific missions.}, archive = {J_COR}, author = {Pedro Marcolin Antunes and Laio Oriel Seman and Eduardo Camponogara}, doi = {10.1016/j.cor.2025.107259}, journal = {Computers & Operations Research}, month = {1}, pages = {107259}, shortjournal = {Comput. Oper. Res.}, title = {A branch-and-price algorithm for energy aware task scheduling of constellations of nanosatellites}, volume = {185}, year = {2026}, }
• (2026). A branch-and-cut algorithm for the pallet-loading vehicle routing problem considering load balance of semi-trailer trucks. COR, 185, 107258. (www)

  @article{ , abstract = {The work presented in this paper is motivated by some practical applications, introducing and solving a new problem called the split delivery vehicle routing problem with time windows and two-dimensional loading constraints considering load balance (2L-SDVRPTW-LB). This problem expands the integrated routing and loading problem, in which the required pallet orders are delivered to customers by a fleet of homogeneous semi-trailer trucks. Firstly, with the goal to minimize the total transportation cost, a mixed-integer linear programming model for the 2L-SDVRPTW-LB is proposed. Secondly, an exact algorithm framework based on the branch-and-cut algorithm is designed to solve the problem. In the algorithm framework, a metaheuristic algorithm is designed to obtain the upper bound. Subsequently, various classes of valid inequalities are introduced to strengthen the formulation, while the constraints ensuring load balance are applied in a lazy manner to optimize computational efficiency. Numerical experiments based on modified benchmark instances demonstrate that the proposed branch-and-cut algorithm effectively solves the 2L-SDVRPTW-LB and its variant of excluding split delivery. This study provides valuable insights for addressing the pallet-loading vehicle routing problem considering load balance of semi-trailer trucks.}, archive = {J_COR}, author = {Xiangbin Xu and Haoxing Ouyang}, doi = {10.1016/j.cor.2025.107258}, journal = {Computers & Operations Research}, month = {1}, pages = {107258}, shortjournal = {Comput. Oper. Res.}, title = {A branch-and-cut algorithm for the pallet-loading vehicle routing problem considering load balance of semi-trailer trucks}, volume = {185}, year = {2026}, }
• (2026). A hybrid memetic metaheuristic for medical staff assignment in major public health emergencies. COR, 185, 107256. (www)

  @article{ , abstract = {During major public health emergencies, effective assignment of medical staff is crucial for saving lives and controlling the spread of epidemics. This work focuses on the assignment of doctors and nurses to hospitals to form treatment groups that carry out patient treatment tasks. We consider the practical constraints of skill types of medical staff and the severity of patients’ conditions and propose a mixed integer programming model with the objective of maximizing demand satisfaction and personnel skill matching. To solve this problem, we introduce a hybrid memetic search algorithm that combines a specialized crossover operator for generating promising offspring solutions and a variable neighborhood search procedure to improve their quality. Computational results demonstrate that our algorithm outperforms the general mixed integer programming solver GUROBI . The key components of the proposed algorithm are experimentally analyzed and managerial insights are derived.}, archive = {J_COR}, author = {Yang Wang and He Zheng and Zequn Wei and Christophe Wilbaut and Saïd Hanafi}, doi = {10.1016/j.cor.2025.107256}, journal = {Computers & Operations Research}, month = {1}, pages = {107256}, shortjournal = {Comput. Oper. Res.}, title = {A hybrid memetic metaheuristic for medical staff assignment in major public health emergencies}, volume = {185}, year = {2026}, }
• (2026). Learning-guided iterated local search for the minmax multiple traveling salesman problem. COR, 185, 107255. (www)

  @article{ , abstract = {The minmax multiple traveling salesman problem involves minimizing the costs of a longest tour among a set of tours. The problem is of great practical interest because it can be used to formulate several real-life applications. To solve this computationally challenging problem, we propose a learning-driven iterated local search approach that combines an effective local search procedure to find high-quality local optimal solutions and a multi-armed bandit algorithm to select removal and insertion operators to escape local optimal traps. Extensive experiments on 77 commonly used benchmark instances show that the algorithm achieves excellent results in terms of solution quality and running time. In particular, it achieves 32 new best results (improved upper bounds) and matches the best-known results for 35 other instances. Additional experiments shed light on the understanding of the algorithm’s constituent elements. Multi-armed bandit selection can be used advantageously in other multi-operator local search algorithms.}, archive = {J_COR}, author = {Pengfei He and Jin-Kao Hao and Jinhui Xia}, doi = {10.1016/j.cor.2025.107255}, journal = {Computers & Operations Research}, month = {1}, pages = {107255}, shortjournal = {Comput. Oper. Res.}, title = {Learning-guided iterated local search for the minmax multiple traveling salesman problem}, volume = {185}, year = {2026}, }
• (2026). Grouping strategies on two-phase methods for bi-objective combinatorial optimization. COR, 185, 107254. (www)

  @article{ , abstract = {Two-phase methods are commonly used to solve bi-objective combinatorial optimization problems. In the first phase, all extreme supported nondominated points are generated through a dichotomic search. This phase also allows the identification of search zones that may contain other nondominated points. The second phase focuses on exploring these search zones to locate the remaining points, which typically accounts for most of the computational cost. Ranking algorithms are frequently employed to explore each zone individually, but this approach leads to redundancies, causing multiple visits to the same solutions. To mitigate these redundancies, we propose several strategies that group adjacent zones, allowing a single run of the ranking algorithm for the entire group. Additionally, we explore an implicit grouping approach based on a new concept of coverage. Our experiments on the Bi-Objective Spanning Tree Problem demonstrate the beneficial impact of these grouping strategies when combined with coverage.}, archive = {J_COR}, author = {Felipe O. Mota and Luís Paquete and Daniel Vanderpooten}, doi = {10.1016/j.cor.2025.107254}, journal = {Computers & Operations Research}, month = {1}, pages = {107254}, shortjournal = {Comput. Oper. Res.}, title = {Grouping strategies on two-phase methods for bi-objective combinatorial optimization}, volume = {185}, year = {2026}, }
• (2026). Drone-aided mobile blood collection problem: A rolling-horizon-based matheuristic. COR, 185, 107253. (www)

  @article{ , abstract = {This study introduces the drone-aided mobile blood collection problem, which integrates mobile blood donation vehicles with drones to improve operations related to the blood collection in urban areas. Each vehicle, carrying multiple drones, travels to several collection sites to conduct blood collection operations within a working day. Drones fly between vehicles to pick up collected blood bags and deliver them to the blood center. This collaborative framework enhances the performances of the collection system and ensures the freshness of collected blood upon arrival to the blood center. We develop a novel mixed-integer linear programming model to optimally synchronize the routes and collection schedules of mobile units and drones to ensure the timely delivery of collected blood to the blood center. We also develop a rolling-horizon-based matheuristic to solve large-scale instances of the problem. This algorithm combines a rolling horizon approach, which divides the problem into manageable subproblems solved sequentially, with a local branching technique that enhances solutions by exploring promising neighborhoods. To evaluate the algorithm’s performance, we conduct a comprehensive computational study. Our results show that the proposed algorithm not only finds better solutions than those obtained by Gurobi but also outperforms other matheuristics, including the rolling horizon, relax-and-fix, and fix-and-optimize algorithms. Finally, we demonstrate the real-life applicability of the problem through a case study in Quebec City, Canada.}, archive = {J_COR}, author = {Amirhossein Abbaszadeh and Hossein Hashemi Doulabi}, doi = {10.1016/j.cor.2025.107253}, journal = {Computers & Operations Research}, month = {1}, pages = {107253}, shortjournal = {Comput. Oper. Res.}, title = {Drone-aided mobile blood collection problem: A rolling-horizon-based matheuristic}, volume = {185}, year = {2026}, }
• (2026). Fair and efficient multi-agent routing for cooperative and autonomous agricultural fleets with implements. COR, 185, 107252. (www)

  @article{ , abstract = {The growing use of autonomous tractor fleets with detachable implements presents complex logistical challenges in agriculture. Current systems often rely on simple heuristics and avoid implement swapping, limiting efficiency. A central challenge is to dynamically coordinate vehicle routing and implement exchanges to enable efficient, low-intervention task execution. Due to high costs, such fleets are owned mainly by large enterprises or cooperatives, where fair task allocation and profit sharing are critical. Addressing both coordination and fairness, in this paper, we introduce the Agricultural Fleet Vehicle Routing Problem with Implements (AFVRPI). We propose a distributed model derived from a centralized formulation also presented in this paper. This model is embedded within a Distributed Multi-Agent System Architecture (DIMASA), where autonomous vehicle agents manage routing and implement use under limited fuel autonomy, while implement agents ensure compatibility and sufficient capacity to meet task demands. Our solution applies systematic egalitarian social welfare optimization to iteratively maximize the profit of the worst-off vehicle, balancing fairness with system efficiency. To enhance scalability, we use column generation in the distributed model, achieving solution quality comparable to the centralized model while significantly reducing computing time. Simulation results on new benchmark instances demonstrate that our distributed multi-agent AFVRPI approach is scalable, efficient, and fair.}, archive = {J_COR}, author = {Aitor López-Sánchez and Marin Lujak and Frédéric Semet and Holger Billhardt}, doi = {10.1016/j.cor.2025.107252}, journal = {Computers & Operations Research}, month = {1}, pages = {107252}, shortjournal = {Comput. Oper. Res.}, title = {Fair and efficient multi-agent routing for cooperative and autonomous agricultural fleets with implements}, volume = {185}, year = {2026}, }
• (2026). Flexible scheduling of customized bus for green mega-events: A distributionally robust optimization approach. COR, 185, 107249. (www)

  @article{ , abstract = {Mega-events such as the Olympics and the World Championships face significant challenges in evacuating large numbers of attendees after the events conclude, which consume substantial transportation resources. Under the global pressure to reduce carbon emissions, energy conservation and emission reduction are increasingly becoming top priorities. This paper focuses on the efficient scheduling of customized buses (CB) after green mega-events, incorporating skip-stop operations and coordinated bus services to minimize energy consumption, fixed and transportation costs, and facilitate the evacuation of attendees, while accounting for practical constraints such as the availability of customized buses, vehicle capacity, time windows, and flow balance. A distributionally robust optimization (DRO) model is developed, using a novel ambiguity set to model uncertain demand via parametric interval-valued fuzzy variables. To ensure computational tractability, the model is reformulated as an integer linear programming model. To address the computational challenges of large-scale instances, an improved variable neighborhood search heuristic is designed by incorporating the reinforcement learning techniques, including the KL-UCB algorithm and a sliding window mechanism. Extensive numerical experiments are conducted to verify the performance of the proposed heuristic. Computational results demonstrate that the proposed DRO model effectively handles uncertainty, offering robust and adaptable solutions. Compared to existing heuristics, the proposed heuristic improves performance by 6.51% on average, and incorporating reinforcement learning into VNS enhances computational efficiency by 4.88% on average. A real-life case study further validates the model, demonstrating that the skip-stop strategy significantly reduces vehicle travel time and enhances overall operational efficiency.}, archive = {J_COR}, author = {Xiaojie An and Xiang Li and Bowen Zhang}, doi = {10.1016/j.cor.2025.107249}, journal = {Computers & Operations Research}, month = {1}, pages = {107249}, shortjournal = {Comput. Oper. Res.}, title = {Flexible scheduling of customized bus for green mega-events: A distributionally robust optimization approach}, volume = {185}, year = {2026}, }
• (2026). Optimal planning of power distribution networks with fault-tolerant configuration. COR, 185, 107248. (www)

  @article{ , abstract = {Power Distribution networks are essential infrastructures that should be designed by satisfying two conflicting requests: cost minimization and reliability. While traditional network planning aimed at radial configurations, which are more similar to the typical working configuration of a network but are not fault-tolerant, modern techniques seek for meshed configurations, since these architectures are more fault-tolerant. Due to the complexity of the problem and the large size of nowadays instances, most of the techniques used for planning are based on heuristic approaches. Thus, they are usually unable to guarantee optimality and not even able to provide an assessment of the distance from the optimal solution. In this work, we address the challenge of planning a fault tolerant network through an exact approach, by introducing innovative Mixed-Integer Linear Programming models designed for the planning of meshed distribution networks with loop-feeder or open-loop topology. Differently from other techniques, our approach simplifies the formulation by avoiding the need for fault scenarios, significantly reducing the computational burden of the optimization problem. The outcomes of our approach are the generation of optimal meshed network, which effectively balance cost and reliability of the electric distribution system. Comprehensive studies on realistic test instances show the advantages of the proposed formulations.}, archive = {J_COR}, author = {Renato Bruni and Alberto Geri and Marco Maccioni and Ludovico Nati}, doi = {10.1016/j.cor.2025.107248}, journal = {Computers & Operations Research}, month = {1}, pages = {107248}, shortjournal = {Comput. Oper. Res.}, title = {Optimal planning of power distribution networks with fault-tolerant configuration}, volume = {185}, year = {2026}, }
• (2026). Centrality measures and opinion dynamics in two-layer networks with replica nodes. COR, 185, 107245. (www)

  @article{ , abstract = {We propose two fast and accurate algorithms to approximate game-theoretic centrality measures and examine connection between centrality measures, network properties, and key performance indicators (consensus time and winning rate) of opinion dynamic processes on such networks. As an example, we consider a Zachary’s karate club as a social network and extend it by adding the second (internal) layer of communication. The internal layer represents the network where individuals can share their real opinions with the close friends. The structures of the external and internal layers may be different. The significant positive correlation between internal graph density and consensus time, and significant negative correlation between centrality of authoritative nodes and consensus time are found. The proposed algorithms are verified by a series of experiments from two aspects: the accuracy and the efficiency. The algorithms are novel and can be considered as a contribution to the network theory independently of opinion dynamics as they can be used to calculate node centrality in any weighted graph.}, archive = {J_COR}, author = {Chi Zhao and Elena Parilina}, doi = {10.1016/j.cor.2025.107245}, journal = {Computers & Operations Research}, month = {1}, pages = {107245}, shortjournal = {Comput. Oper. Res.}, title = {Centrality measures and opinion dynamics in two-layer networks with replica nodes}, volume = {185}, year = {2026}, }
• (2026). AI automatic decision in newsvendor model with nash bargaining fairness concern. COR, 185, 107227. (www)

  @article{ , abstract = {This paper investigates the impact of artificial intelligence (AI) automatic ordering and producing decisions on fairness-concerned supply chains under the newsvendor model. We develop a dyadic supply chain model in which the manufacturer acts as the Stackelberg leader while the retailer serves as the follower in a push supply chain. In contrast, their roles are switched in a pull supply chain. We assume that only human decision-making leads to decision regret behavior, whereas AI-automated decision-making does not. Without adopting AI, our results show that fairness concern does not necessarily lead to a decreasing quantity in ordering or producing, which is different from most previous studies. Different from the prior findings, our work reveals that in binding equilibrium, if fairness concerns are considered, the order quantity will decrease, while in non-binding equilibrium, the order quantity may not necessarily be less than the previous results. Interestingly, when decision regret bias is considered for fairness-concerned decision-makers, we can obtain quantity coordination solutions for supply chains under specific conditions. With adopting AI, our results show that increasing fairness concerns are beneficial for improving the follower’s profit while at the expense of sacrificing the leader’s profit margins, while the leader can only benefit from AI adoption when the decision regret bias of the follower is relatively high. It is noteworthy that under certain conditions, AI automation may negatively impact the profits of both push and pull decentralized supply chains. For instance, in low-margin profit scenarios where decision-makers exhibit moderate regret bias and fairness concerns, such effects can emerge. This indicates that under specific circumstances, the human behavioral factors — regret bias and fairness concerns — may sometimes enhance the performance of decentralized supply chain members. Our research findings provide significant practical implications for the adoption of AI-automated decision-making in real-world supply chains.}, archive = {J_COR}, author = {Rui Hou and Yishen Cen and Jianxin Chen}, doi = {10.1016/j.cor.2025.107227}, journal = {Computers & Operations Research}, month = {1}, pages = {107227}, shortjournal = {Comput. Oper. Res.}, title = {AI automatic decision in newsvendor model with nash bargaining fairness concern}, volume = {185}, year = {2026}, }
• (2026). Green horizons: Sustainable global logistics in dynamic supply chain management. COR, 185, 107226. (www)

  @article{ , abstract = {Supply chain management in a global scale involves addressing numerous uncertainties, from demand fluctuations to unforeseen disruptions. Developing advanced solution approaches is critical to manage such complexities and ensure resilience. This study presents a multi-stage stochastic–dynamic model for the global supply chain, incorporating hedging policies. The aim is to identify optimal order scheduling for bill of materials, production planning, and inventory management across warehouses (i.e., materials and finished products). Due to the dynamic nature of the global supply chain (e.g., demand fluctuations, disruptions, and lead time), a multi-stage stochastic model is developed for the stochastic–dynamic supply chain network. To address dynamic factors of real-world global supply chain, an accelerated parallel stochastic dual dynamic integer programming (SDDiP) approach is proposed to deal with disruptions (e.g., political unrest, natural disasters, and pandemics), enhancing supply chain resiliency. To validate the proposed parallel SDDiP , various scenarios with different sizes are generated using the case study and compared to the SDDiP with Benders cuts and integrated stage-wise Lagrangian dual cut ( SWLDC ) (i.e., SDDiP-SWLDC ). According to the obtained results, the proposed parallel node strategy for accelerated SDDiP consistently outperforms the basic stochastic dual dynamic programming (SDDP) and demonstrated robust CPU scalability. Evaluation across various scenario sizes shows stochastic dual dynamic integer programming-mixed integer rounding cuts ( SDDiP-MIR ) achieving faster computation and a smaller 7% optimality gap compared to SDDiP-SWLDC and SDDiP in large-size instances, highlighting its superior performance in complex supply chain settings.}, archive = {J_COR}, author = {Mahsa Mohammadi and Babak Mohamadpour Tosarkani}, doi = {10.1016/j.cor.2025.107226}, journal = {Computers & Operations Research}, month = {1}, pages = {107226}, shortjournal = {Comput. Oper. Res.}, title = {Green horizons: Sustainable global logistics in dynamic supply chain management}, volume = {185}, year = {2026}, }

CSDA - 8

• (2026). Fast and efficient causal inference in large-scale data via subsampling and projection calibration. CSDA, 214, 108281. (www)

  @article{ , abstract = {Estimating the average treatment effect in large-scale datasets faces significant computational and storage challenges. Subsampling has emerged as a critical strategy to mitigate these issues. This paper proposes a novel subsampling method that builds on the G-estimation method offering the double robustness property. The proposed method uses a small subset of data to estimate computationally complex nuisance parameters, while leveraging the full dataset for the computationally simple final estimation. To ensure that the resulting estimator remains first-order insensitive to variations in nuisance parameters, a projection approach is introduced to optimize the estimation of the outcome regression function and treatment regression function such that the Neyman orthogonality conditions are satisfied. It is shown that the resulting estimator is asymptotically normal and achieves the same convergence rate as the full data-based estimator when either the treatment or the outcome models is correctly specified. Additionally, when both models are correctly specified, the proposed estimator achieves the same asymptotic variance as the full data-based estimator. The finite sample performance of the proposed method is demonstrated through simulation studies and an application to birth data, comprising over 30 million observations collected over the past eight years. Numerical results indicate that the proposed estimator is nearly as computationally efficient as the uniform subsampling estimator, while achieving similar estimation efficiency to the full data-based G-estimator.}, archive = {J_CSDA}, author = {Miaomiao Su}, doi = {10.1016/j.csda.2025.108281}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108281}, shortjournal = {Comput. Stat. Data Anal.}, title = {Fast and efficient causal inference in large-scale data via subsampling and projection calibration}, volume = {214}, year = {2026}, }
• (2026). An algorithm for estimating threshold boundary regression models. CSDA, 214, 108274. (www)

  @article{ , abstract = {This paper presents an innovative iterative two-stage algorithm designed for estimating threshold boundary regression (TBR) models. By transforming the non-differentiable least-squares (LS) problem inherent in fitting TBR models into an optimization framework, our algorithm combines the optimization of a weighted classification error function for the threshold model with obtaining LS estimators for regression models. To improve the efficiency and flexibility of TBR model estimation, we integrate the weighted support vector machine (WSVM) as a surrogate method for solving the weighted classification problem. The TBR-WSVM algorithm offers several key advantages over recently developed methods: it eliminates pre-specification requirements for threshold parameters, accommodates flexible estimation of nonlinear threshold boundaries, and streamlines the estimation process. We conducted several simulation studies to illustrate the finite-sample performance of TBR-WSVM. Finally, we demonstrate the practical applicability of the TBR model through a real data analysis.}, archive = {J_CSDA}, author = {Chih-Hao Chang and Takeshi Emura and Shih-Feng Huang}, doi = {10.1016/j.csda.2025.108274}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108274}, shortjournal = {Comput. Stat. Data Anal.}, title = {An algorithm for estimating threshold boundary regression models}, volume = {214}, year = {2026}, }
• (2026). Rate accelerated inference for integrals of multivariate random functions. CSDA, 214, 108273. (www)

  @article{ , abstract = {The computation of integrals is a fundamental task in the analysis of functional data, where the data are typically considered as random elements in a space of squared integrable functions. Effective unbiased estimation and inference procedures are proposed for integrals of uni- and multivariate random functions. Applications to key problems in functional data analysis involving random design points are examined and illustrated. In the absence of noise, the proposed estimates converge faster than the sample mean and standard numerical integration algorithms. The estimator also supports effective inference by generally providing better coverage with shorter confidence and prediction intervals in both noisy and noiseless settings.}, archive = {J_CSDA}, author = {Valentin Patilea and Sunny G․ W․ Wang}, doi = {10.1016/j.csda.2025.108273}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108273}, shortjournal = {Comput. Stat. Data Anal.}, title = {Rate accelerated inference for integrals of multivariate random functions}, volume = {214}, year = {2026}, }
• (2026). Robust selection of the number of change-points via FDR control. CSDA, 214, 108272. (www)

  @article{ , abstract = {Robust quantification of uncertainty regarding the number of change-points presents a significant challenge in data analysis, particularly when employing false discovery rate (FDR) control techniques. Emphasizing the detection of genuine signals while controlling false positives is crucial, especially for identifying shifts in location parameters within flexible distributions. Traditional parametric methods often exhibit sensitivity to outliers and heavy-tailed data. Addressing this limitation, a robust method accommodating diverse data structures is proposed. The approach constructs component-wise sign-based statistics. Leveraging the global symmetry inherent in these statistics enables the derivation of data-driven thresholds suitable for multiple testing scenarios. Method development occurs within the framework of U-statistics, which naturally encompasses existing cumulative sum-based procedures. Theoretical guarantees establish FDR control for the component-wise sign-based method under mild assumptions. Demonstrations of effectiveness utilize simulations with synthetic data and analyses of real data.}, archive = {J_CSDA}, author = {Hui Chen and Chengde Qian and Qin Zhou}, doi = {10.1016/j.csda.2025.108272}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108272}, shortjournal = {Comput. Stat. Data Anal.}, title = {Robust selection of the number of change-points via FDR control}, volume = {214}, year = {2026}, }
• (2026). Kernel density estimation with a markov chain monte carlo sample. CSDA, 214, 108271. (www)

  @article{ , abstract = {Bayesian inference relies on the posterior distribution, which is often estimated with a Markov chain Monte Carlo sampler. The sampler produces a dependent stream of variates from the limiting distribution of the Markov chain, the posterior distribution. When one wishes to display the estimated posterior density, a natural choice is the histogram. However, abundant literature has shown that the kernel density estimator is more accurate than the histogram in terms of mean integrated squared error for an i.i.d. sample. With this as motivation, a kernel density estimation method is proposed that is appropriate for the dependence in the Markov chain Monte Carlo output. To account for the dependence, the cross-validation criterion is modified to select the bandwidth in standard kernel density estimation approaches. A data-driven adjustment to the biased cross-validation method is suggested with introducing the integrated autocorrelation time of the kernel. The convergence of the modified bandwidth to the optimal bandwidth is shown by adapting theorems from the time series literature. Simulation studies show that the proposed method finds the bandwidth close to the optimal value, while standard methods lead to smaller bandwidths under Markov chain samples and hence to undersmoothed density estimates. A study with real data shows that the proposed method has a considerably smaller integrated mean squared error than standard methods. The R package KDEmcmc to implement the suggested algorithm is available on the Comprehensive R Archive Network.}, archive = {J_CSDA}, author = {Hang J. Kim and Steven N. MacEachern and Young Min Kim and Yoonsuh Jung}, doi = {10.1016/j.csda.2025.108271}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108271}, shortjournal = {Comput. Stat. Data Anal.}, title = {Kernel density estimation with a markov chain monte carlo sample}, volume = {214}, year = {2026}, }
• (2026). Measure selection for functional linear model. CSDA, 214, 108270. (www)

  @article{ , abstract = {Advancements in modern science have led to an increased prevalence of functional data, which are usually viewed as elements of the space of square-integrable functions L 2 . Core methods in functional data analysis, such as functional principal component analysis, are typically grounded in the Hilbert structure of L 2 and rely on inner products based on integrals with respect to the Lebesgue measure over a fixed domain. A more flexible framework is proposed, where the measure can be arbitrary, allowing natural extensions to unbounded domains and prompting the question of optimal measure choice. Specifically, a novel functional linear model is introduced that incorporates a data-adaptive choice of the measure that defines the space, alongside an enhanced function principal component analysis. Selecting a good measure can improve the model’s predictive performance, especially when the underlying processes are not well-represented when adopting the default Lebesgue measure. Simulations, as well as applications to COVID-19 data and the National Health and Nutrition Examination Survey data, show that the proposed approach consistently outperforms the conventional functional linear model.}, archive = {J_CSDA}, author = {Su I Iao and Hans-Georg Müller}, doi = {10.1016/j.csda.2025.108270}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108270}, shortjournal = {Comput. Stat. Data Anal.}, title = {Measure selection for functional linear model}, volume = {214}, year = {2026}, }
• (2026). Overview of normal-reference tests for high-dimensional means with implementation in the r package ‘HDNRA’. CSDA, 214, 108269. (www)

  @article{ , abstract = {The challenge of testing for equal mean vectors in high-dimensional data poses significant difficulties in statistical inference. Much of the existing literature introduces methods that often rely on stringent regularity conditions for the underlying covariance matrices, enabling asymptotic normality of test statistics. However, this can lead to complications in controlling test size. To address these issues, a new set of tests has emerged, leveraging the normal-reference approach to improve reliability. The latest normal-reference methods for testing equality of mean vectors in high-dimensional samples, potentially with differing covariance structures, are reviewed. The theoretical underpinnings of these tests are revisited, providing a new unified justification for the validity of centralized L 2 -norm-based normal-reference tests (NRTs) by deriving the convergence rate of the distance between the null distribution of the test statistic and its corresponding normal-reference distribution. To facilitate practical application, an R package, HDNRA , is introduced, implementing these NRTs and extending beyond the two-sample problem to accommodate general linear hypothesis testing (GLHT). The package, designed with user-friendliness in mind, achieves efficient computation through a core implemented in C++ using Rcpp , OpenMP , and RcppArmadillo . Examples with real datasets are included, showcasing the application of various tests and providing insights into their practical utility.}, archive = {J_CSDA}, author = {Pengfei Wang and Tianming Zhu and Jin-Ting Zhang}, doi = {10.1016/j.csda.2025.108269}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108269}, shortjournal = {Comput. Stat. Data Anal.}, title = {Overview of normal-reference tests for high-dimensional means with implementation in the r package ‘HDNRA’}, volume = {214}, year = {2026}, }
• (2026). High-dimensional subgroup functional quantile regression with panel and dependent data. CSDA, 214, 108268. (www)

  @article{ , abstract = {High-dimensional additive functional partial linear single-index quantile regression with high-dimensional parameters under subgroup panel data is investigated. Based on spline-based approach, we construct oracle estimators of the unknown parameter and functions, and discuss their consistency with rates and asymptotic normality under α -mixing assumptions. A penalized estimation method by using the SCAD technique is introduced to estimate the additive functions and parameter, enabling variable selection and automatic identification of the number of groups. Hypothesis testing for the parameter is also considered, and the asymptotic distributions of the restricted estimators and the test statistic are derived under both the null and local alternative hypotheses. Simulation studies and real data analysis are conducted to verify the validity of the proposed methods and applications.}, archive = {J_CSDA}, author = {Xiao-Ge Yu and Han-Ying Liang}, doi = {10.1016/j.csda.2025.108268}, journal = {Computational Statistics & Data Analysis}, month = {2}, pages = {108268}, shortjournal = {Comput. Stat. Data Anal.}, title = {High-dimensional subgroup functional quantile regression with panel and dependent data}, volume = {214}, year = {2026}, }

CVIU - 10

• (2025). KD-mamba: Selective state space models with knowledge distillation for trajectory prediction. CVIU, 261, 104499. (www)

  @article{ , abstract = {Trajectory prediction is a key component of intelligent mobility systems and human–robot interaction. The inherently stochastic nature of human behavior, coupled with external environmental influences, poses significant challenges for long-term prediction. However, existing approaches struggle to effectively model spatial interactions and accurately predict long-term destinations, while their high computational demands limit real-world applicability. To address these limitations, this paper presents KD-Mamba, the Selective State Space Models with Knowledge Distillation for trajectory prediction. The model incorporates the U-CMamba module, which features a U-shaped encoder–decoder architecture. By integrating convolutional neural networks (CNN) with the Mamba mechanism, this module effectively captures local spatial interactions and global contextual information of human motion patterns. Subsequently, we introduce a Bi-Mamba module, which captures long-term dependencies in human movement, ensuring a more accurate representation of trajectory dynamics. Knowledge distillation strengthens both modules by facilitating knowledge transfer across diverse scenarios. Compared to transformer-based approaches, KD-Mamba reduces computational complexity from quadratic to linear. Extensive experimental results from two real-world trajectory datasets indicate that KD-Mamba outperforms the existing mainstream baselines. The proposed method provides insights into the application of trajectory prediction in human-in-the-loop assistive systems.}, archive = {J_CVIU}, author = {Shaokang Cheng and Sourav Das and Shiru Qu and Lamberto Ballan}, doi = {10.1016/j.cviu.2025.104499}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104499}, shortjournal = {Comput. Vis. Image Understanding}, title = {KD-mamba: Selective state space models with knowledge distillation for trajectory prediction}, volume = {261}, year = {2025}, }
• (2025). Two-stage attribute-guided dual attention network for fine-grained fashion retrieval. CVIU, 261, 104497. (www)

  @article{ , abstract = {Fine-grained clothing retrieval is essential for intelligent shopping and personalized recommendation systems. However, conventional methods often fail to capture subtle attribute variations. This paper proposes a novel two-stage attribute-guided dual attention network. The network combines global and local feature extraction with Attribute-aware Multi-Scale Spatial Attention (AMSA) and Attribute-guided Dynamic Channel Attention (ADCA). AMSA captures attribute-specific spatial details at multiple scales, while ADCA dynamically adjusts channel importance based on attribute embeddings, enabling precise attribute-level similarity modeling. A multi-level joint loss function further optimizes both global and local representations and enhances feature alignment. Experiments on FashionAI and the self-built FGDress dataset show that the proposed method achieves mAP scores of 66.01% and 73.98%, respectively, outperforming baseline approaches. Attribute-level analysis confirms robust recognition of both well-defined and challenging attributes. These results validate the practicality and generalizability of the proposed framework, with promising applications in personalized recommendation, fashion trend analysis, and design evaluation.}, archive = {J_CVIU}, author = {Bo Pan and Jun Xiang and Ning Zhang and Ruru Pan}, doi = {10.1016/j.cviu.2025.104497}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104497}, shortjournal = {Comput. Vis. Image Understanding}, title = {Two-stage attribute-guided dual attention network for fine-grained fashion retrieval}, volume = {261}, year = {2025}, }
• (2025). JSF: A joint spatial-frequency domain network for low-light image enhancement. CVIU, 261, 104496. (www)

  @article{ , abstract = {The enhancement of low-light images remains a prominent focus in the field of image processing. The degree of lightness significantly influences vision-based intelligent recognition and analysis. Departing from conventional methods, this paper proposes an innovative joint spatial-frequency domain network for low-light image enhancement, referred to as JSF. In the spatial domain, brightness is optimized through the amalgamation of global and local information. In the frequency domain, noise is reduced and details are amplified using Fourier Transformation to carry out amplitude and phase enhancement. Additionally, the enhanced results from the aforementioned domains are fused by linear and nonlinear stretching. To validate the effectiveness of JSF, this paper presents both qualitative and quantitative comparison results, demonstrating its superiority over several existing state-of-the-art methods.}, archive = {J_CVIU}, author = {Yahong Wu and Feng Liu and Rong Wang}, doi = {10.1016/j.cviu.2025.104496}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104496}, shortjournal = {Comput. Vis. Image Understanding}, title = {JSF: A joint spatial-frequency domain network for low-light image enhancement}, volume = {261}, year = {2025}, }
• (2025). FCNet: A feature complementary network for nighttime flare removal. CVIU, 261, 104495. (www)

  @article{ , abstract = {Nighttime image flare removal is a very challenging task due to the presence of various types of unfavorable degrading effects, including glare, shimmer, streak and saturated blobs. Most of the existing methods focus on the spatial domain and limited perception field, resulting in incomplete flare removal and severe artifacts. To address these challenges, we propose a two-stage feature complementary network for nighttime flare removal, which is used for flare perception and removal, respectively. In the first stage, a Spatial-Frequency Complementary Module (SFCM) is designed to perceive the flare region from different domains to get a mask of the flare. In the second stage, the flare mask and image are fed into the Spatial-Frequency Complementary Gating Module (SFCGM) to preserve the background information, while removing the flares from different angles and restoring the detailed features. Finally the flare and non-flare regions are modeled by the Flare Interactive Module (FIM) to refine the flare regions at a fine-grained level to suppress the artifact problem. Extensive experiments on Flare 7K++ validate the superiority of the proposed approach over state-of-the-arts, both qualitatively and quantitatively.}, archive = {J_CVIU}, author = {Kejing Qi and Bo Wang and Chongyi Li}, doi = {10.1016/j.cviu.2025.104495}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104495}, shortjournal = {Comput. Vis. Image Understanding}, title = {FCNet: A feature complementary network for nighttime flare removal}, volume = {261}, year = {2025}, }
• (2025). GRE-net: A forgery image detection framework based on gradient feature and reconstruction error. CVIU, 261, 104494. (www)

  @article{ , abstract = {With the continuous technological breakthroughs in Generative Adversarial Networks (GANs) and diffusion models, remarkable progress has been achieved in the field of image generation. These technologies enable the creation of highly realistic images, thereby intensifying the risk of spreading fake information. However, traditional image detectors face a growing challenge of inadequate generalization capabilities when confronted with images generated by models that were not included during the training phase. To tackle this challenge, we introduce a novel detection framework, named GRE-Net (Network integrating Gradient and Reconstruction Error), which extracts gradient feature through the DPG module and calculates the reconstruction error utilizing the DIRE method. By integrating these two aspects into a comprehensive feature representation, GRE-Net effectively detects the authenticity of images. Specifically, we devise a dual-branch model that leverages the proposed DPG (Discriminator of ProjectedGAN to extract Gradient) module to extract gradient feature from images and concurrently employs the DIRE (DIffusion Reconstruction Error) method to obtain the diffusion reconstruction error of images. By fusing the features extracted from these two modules as a universal representation, we describe the artifacts produced by generative models, crafting a comprehensive detector capable of identifying both GAN-generated and diffusion model-generated images. Notably, the DPG approach utilizes the discriminator of ProjectedGAN as an intermediary bridge, mapping all data into the gradient domain. This transformation process effectively captures the intrinsic feature differences during the image generation process. Subsequently, the gradient feature are fed into a classifier to achieve efficient discrimination between authentic and fake images. To validate the efficacy of our proposed detector, we conducted evaluations on a dataset comprising images generated by ten diverse diffusion models and GANs. Extensive experiments demonstrate that our detector exhibits stronger generalization capabilities and higher robustness, rendering it suitable for real-world generated image detection tasks.}, archive = {J_CVIU}, author = {Wenqing Wu and Xinyi Shi and Jinghai Ai and Xiaodong Wang}, doi = {10.1016/j.cviu.2025.104494}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104494}, shortjournal = {Comput. Vis. Image Understanding}, title = {GRE-net: A forgery image detection framework based on gradient feature and reconstruction error}, volume = {261}, year = {2025}, }
• (2025). Mitigating forgetting in the adaptation of CLIP for few-shot classification. CVIU, 261, 104493. (www)

  @article{ , abstract = {Adapter-style efficient transfer learning has demonstrated outstanding performance in fine-tuning vision-language models, especially in scenarios with limited data. However, existing methods fail to effectively balance the prior knowledge acquired during the pre-training process and the training samples. To address this problem, we propose a method called Mitigating Forgetting in the Adaptation (MiFA) of CLIP. MiFA first employs class prototypes to represent the most prominent features of a class, and these prototypes provide a robust initialization for the classifier. To overcome the forgetting of prior knowledge, MiFA then leverages a memory module that retains the initial parameters and the parameters of training history by creating a memory weight through momentum. The weight is used to initialize a new classification layer, which, along with the original layer, guides each other to balance prior knowledge and feature adaptation. Similarly, in the text processing branch, a parallel initialization strategy is adopted to ensure that the model’s performance is improved. Text features are employed to initialize a text classification layer, and CLIP logits help prevent excessive forgetting of useful text information. Extensive experiments have demonstrated the effectiveness of our method.}, archive = {J_CVIU}, author = {Jiale Cao and Yuanheng Liu and Zhong Ji and Jingren Liu and Aiping Yang and Yanwei Pang}, doi = {10.1016/j.cviu.2025.104493}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104493}, shortjournal = {Comput. Vis. Image Understanding}, title = {Mitigating forgetting in the adaptation of CLIP for few-shot classification}, volume = {261}, year = {2025}, }
• (2025). Joint multi-dimensional dynamic attention and transformer for general image restoration. CVIU, 261, 104491. (www)

  @article{ , abstract = {Outdoor images often suffer from severe degradation due to rain, haze, and noise, impairing image quality and challenging high-level tasks. Current image restoration methods struggle to handle complex degradation while maintaining efficiency. This paper introduces a novel image restoration architecture that combines multi-dimensional dynamic attention and self-attention within a U-Net framework. To leverage the global modeling capabilities of transformers and the local modeling capabilities of convolutions, we integrate sole CNNs in the encoder–decoder and sole transformers in the latent layer. Additionally, we design convolutional kernels with selected multi-dimensional dynamic attention to capture diverse degraded inputs efficiently. A transformer block with transposed self-attention further enhances global feature extraction while maintaining efficiency. Extensive experiments demonstrate that our method achieves a better balance between performance and computational complexity across five image restoration tasks: deraining, deblurring, denoising, dehazing, and enhancement, as well as superior performance for high-level vision tasks. The source code will be available at https://github.com/House-yuyu/MDDA-former .}, archive = {J_CVIU}, author = {Huan Zhang and Xu Zhang and Nian Cai and Jianglei Di and Yun Zhang}, doi = {10.1016/j.cviu.2025.104491}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104491}, shortjournal = {Comput. Vis. Image Understanding}, title = {Joint multi-dimensional dynamic attention and transformer for general image restoration}, volume = {261}, year = {2025}, }
• (2025). LAM-YOLO: Drones-based small object detection on lighting-occlusion attention mechanism YOLO. CVIU, 261, 104489. (www)

  @article{ , abstract = {Drone-based target detection presents inherent challenges, including the high density and overlap of targets in drone images, as well as the blurriness of targets under varying lighting conditions, which complicates accurate identification. Traditional methods often struggle to detect numerous small, densely packed targets against complex backgrounds. To address these challenges, we propose LAM-YOLO, an object detection model specifically designed for drone-based applications. First, we introduce a light-occlusion attention mechanism to enhance the visibility of small targets under diverse lighting conditions. Additionally, we incorporate Involution modules to improve feature layer interactions. Second, we employ an improved SIB-IoU as the regression loss function to accelerate model convergence and enhance localization accuracy. Finally, we implement a novel detection strategy by introducing two auxiliary detection heads to better identify smaller-scale targets. Our quantitative results demonstrate that LAM-YOLO outperforms methods such as Faster R-CNN, YOLOv11, and YOLOv12 in terms of mAP@0.5 and mAP@0.5:0.95 on the VisDrone2019 public dataset. Compared to the original YOLOv8, the average precision increases by 7.1%. Additionally, the proposed SIB-IoU loss function not only accelerates convergence speed during training but also improves average precision compared to the traditional loss function.}, archive = {J_CVIU}, author = {Yuchen Zheng and Yuxin Jing and Jufeng Zhao and Guangmang Cui}, doi = {10.1016/j.cviu.2025.104489}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104489}, shortjournal = {Comput. Vis. Image Understanding}, title = {LAM-YOLO: Drones-based small object detection on lighting-occlusion attention mechanism YOLO}, volume = {261}, year = {2025}, }
• (2025). Camera pose in SfT and NRSfM under isometric and weaker deformation models. CVIU, 261, 104488. (www)

  @article{ , abstract = {Camera pose is a very natural concept in 3D vision in the rigid setting. It is however much more difficult to work with in deformable settings. Consequently, numerous deformable reconstruction methods simply ignore camera pose. We analyse the concept of pose in deformable settings and prove that it is unconstrained with the existing formulations, properly justifying the existing pose-less methods reconstructing structure only. We explain this result intuitively by the impossibility to define an intrinsic coordinate frame to a general deforming object. The proposed analysis uses the isometric deformation model and extends to the weaker models including conformality and equiareality We propose a novel prior to rescue camera pose estimation in deformable settings, which attributes the deforming object’s dominant rigid-body motion to the camera. We show that adding this prior to any existing formulation fully constrains camera pose and leads to elegant two-step solution methods, involving deformable structure reconstruction using a base method in the first step, and absolute orientation or Procrustes analysis in the second step. We derive the proposed approach for the template-based and template-less settings, respectively implemented using Shape-from-Template (SfT) and Non-Rigid Structure-from-Motion (NRSfM) as base methods and validate them experimentally, showing that the computed pose is qualitatively and quantitatively plausible.}, archive = {J_CVIU}, author = {Adrien Bartoli and Agniva Sengupta}, doi = {10.1016/j.cviu.2025.104488}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104488}, shortjournal = {Comput. Vis. Image Understanding}, title = {Camera pose in SfT and NRSfM under isometric and weaker deformation models}, volume = {261}, year = {2025}, }
• (2025). Bi-granularity balance learning for long-tailed image classification. CVIU, 261, 104469. (www)

  @article{ , abstract = {In long-tailed datasets, the training of deep neural network-based models faces challenges, where the model may become biased towards the head classes with abundant training data, resulting in poor performance on tail classes with limited samples. Most current methods employ contrastive learning to learn more balanced representations by finding the class center. However, these methods use class centers to address local imbalance within a mini-batch, they overlook the global imbalance between batches throughout an epoch, caused by the long-tailed distribution of the dataset. In this paper, we propose bi-granularity balance learning to address the two-layer imbalance. We decouple the attraction–repulsion term in contrastive loss into two independent components: global and local balance. The global balance component focuses on capturing semantic information from different perspectives of the image and shifting learning attention from the head classes to the tail classes in the global perspective. The local balance component aims to learn inter-class separability from the local perspective. The proposed method efficiently learns the intra-class compactness and inter-class separability in long-tailed model training and improves the performance of the long-tailed model. Experimental results show that the proposed method achieves competitive performance on long-tailed benchmarks such as CIFAR-10/100-LT, TinyImageNet-LT, and iNaturalist 2018.}, archive = {J_CVIU}, author = {Ning Ren and Xiaosong Li and Yanxia Wu and Yan Fu}, doi = {10.1016/j.cviu.2025.104469}, journal = {Computer Vision and Image Understanding}, month = {11}, pages = {104469}, shortjournal = {Comput. Vis. Image Understanding}, title = {Bi-granularity balance learning for long-tailed image classification}, volume = {261}, year = {2025}, }

DISOPT - 3

• (2025). Computational aspects of lifted cover inequalities for knapsacks with few different weights. DISOPT, 58, 100912. (www)

  @article{ , abstract = {Cutting planes are frequently used for solving integer programs. A common strategy is to derive cutting planes from building blocks or a substructure of the integer program. In this paper, we focus on knapsack constraints that arise from single row relaxations. Among the most popular classes derived from knapsack constraints are lifted minimal cover inequalities. The separation problem for these inequalities is NP-hard though, and one usually separates them heuristically, therefore not fully exploiting their potential. For many benchmarking instances however, it turns out that many knapsack constraints only have few different coefficients. This motivates the concept of sparse knapsacks where the number of different coefficients is a small constant, independent of the number of variables present. For such knapsacks, we observe that there are only polynomially many different classes of structurally equivalent minimal covers. This opens the door to specialized techniques for using lifted minimal cover inequalities. In this article we will discuss two such techniques, which are based on specialized sorting methods. On the one hand, we present new separation routines that separate equivalence classes of inequalities rather than individual inequalities. On the other hand, we derive compact extended formulations that express all lifted minimal cover inequalities by means of a polynomial number of constraints. These extended formulations are based on tailored sorting networks that express our separation algorithm by linear inequalities. We conclude the article by a numerical investigation of the different techniques for popular benchmarking instances.}, archive = {J_DISOPT}, author = {Christopher Hojny and Cédric Roy}, doi = {10.1016/j.disopt.2025.100912}, journal = {Discrete Optimization}, month = {11}, pages = {100912}, shortjournal = {Discret. Optim.}, title = {Computational aspects of lifted cover inequalities for knapsacks with few different weights}, volume = {58}, year = {2025}, }
• (2025). An improved bound for the price of anarchy for related machine scheduling. DISOPT, 58, 100911. (www)

  @article{ , abstract = {In this paper, we introduce an improved upper bound for the efficiency of Nash equilibria in utilitarian scheduling games on related machines. The machines have varying speeds and adhere to the shortest processing time first policy. The goal of each job is to minimize its completion time, while the social objective is to minimize the sum of completion times. Our main finding establishes an upper bound of 2 − 1 / ( 4 m − 2 ) on the price of anarchy for the general case of m machines. We improve this bound to 3/2 for the case of two machines, and to 2 − 1 / ( 2 m ) for the general case of m machines when the machines have divisible speeds, i.e., if the speed of each machine is divisible by the speed of any slower machine.}, archive = {J_DISOPT}, author = {André Berger and Arman Rouhani and Marc Schröder}, doi = {10.1016/j.disopt.2025.100911}, journal = {Discrete Optimization}, month = {11}, pages = {100911}, shortjournal = {Discret. Optim.}, title = {An improved bound for the price of anarchy for related machine scheduling}, volume = {58}, year = {2025}, }
• (2025). On the circuit diameter conjecture for counterexamples to the hirsch conjecture. DISOPT, 58, 100910. (www)

  @article{ , abstract = {Circuit diameters of polyhedra are a fundamental tool for studying the complexity of circuit augmentation schemes for linear programming and for finding lower bounds on combinatorial diameters. The main open problem in this area is the circuit diameter conjecture, the analogue of the Hirsch conjecture in the circuit setting. A natural question is whether the well-known counterexamples to the Hirsch conjecture carry over. Previously, Stephen and Yusun showed that the Klee-Walkup counterexample to the unbounded Hirsch conjecture does not transfer to the circuit setting. Our main contribution is to show that the original counterexamples for other variants, using monotone walks or for bounded polytopes, also do not transfer. A challenge lies in the dependence of circuit diameters on the specific realization of a polyhedron. We discuss for which realizations, in addition to the original ones from the literature, our tools resolve this question. Our results rely on new observations on structural properties of these counterexamples. To analyze the bounded case, we exploit the geometry of certain 2-faces of the polytopes underlying all known bounded Hirsch counterexamples in Santos’ work. For Todd’s monotone Hirsch counterexample, we study linear programs on spindles and prove sufficient conditions for short monotone circuit walks to exist. We then enumerate all linear programs over Todd’s polytope and find four new orientations that contradict the monotone Hirsch conjecture, while the remaining 7107 satisfy the bound. The conclusion then follows by applying these sufficient conditions to Todd’s counterexample.}, archive = {J_DISOPT}, author = {Alexander E. Black and Steffen Borgwardt and Matthias Brugger}, doi = {10.1016/j.disopt.2025.100910}, journal = {Discrete Optimization}, month = {11}, pages = {100910}, shortjournal = {Discret. Optim.}, title = {On the circuit diameter conjecture for counterexamples to the hirsch conjecture}, volume = {58}, year = {2025}, }

DKE - 14

• (2026). ASF: A novel associative scoring function for embedded knowledge graph reasoning. DKE, 161, 102511. (www)

  @article{ , abstract = {One of the most important tools for knowledge management is the Knowledge Graph (KG), a multi-relational graph that depicts rich factual information across entities. A KG represents entities as nodes and relations as edges, with each edge represented by a triplet: (head entity, relation, tail entity). The Scoring Function (SF) in a KG quantifies the plausibility of these triplets and is often derived from KG embeddings. However, due to the distinct relational patterns across KGs, an SF that performs well on one KG might fail on another, making the design of optimal SFs a challenging task. This study introduces the concept of an Associative Scoring Function (ASF), which leverages Association Rule Mining (ARM) to discover and incorporate patterns and characteristics of symmetric, asymmetric, inverse, and other relational types within embedded KGs. The ARM technique in ASF uses the FP-Growth algorithm to extract meaningful associations, which is enhanced further through hyperparameter tuning. Extensive experiments on benchmark datasets demonstrate that ASF is KG-independent and performs better than state-of-the-art SFs. These results highlight ASF's potential to generalize across diverse KGs, offering a significant advancement in the KG link prediction task.}, archive = {J_DKE}, author = {MVPT Lakshika and HA Caldera}, doi = {10.1016/j.datak.2025.102511}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102511}, shortjournal = {Data Knowl. Eng.}, title = {ASF: A novel associative scoring function for embedded knowledge graph reasoning}, volume = {161}, year = {2026}, }
• (2026). A graph-based model for semantic textual similarity measurement. DKE, 161, 102509. (www)

  @article{ , abstract = {Measuring semantic similarity between sentence pairs is a fundamental problem in Natural Language Processing with applications in various domains, including machine translation, speech recognition, automatic question answering, and text summarization. Despite its significance, accurately assessing semantic similarity remains a challenging task, particularly for underrepresented languages such as Vietnamese. Existing methods have yet to fully leverage the unique linguistic characteristics of Vietnamese for semantic similarity measurement. To address this limitation, we propose GBNet-STS (Graph-Based Network for Semantic Textual Similarity), a novel framework for measuring the semantic similarity of Vietnamese sentence pairs. GBNet-STS integrates lexical-grammatical similarity scores and distributional semantic similarity scores within a multi-layered graph-based model. By capturing different semantic perspectives through multiple interconnected layers, our approach provides a more comprehensive and robust similarity estimation. Experimental results demonstrate that GBNet-STS outperforms traditional methods, achieving state-of-the-art performance in Vietnamese semantic similarity tasks.}, archive = {J_DKE}, author = {Van-Tan Bui and Quang-Minh Nguyen and Van-Vinh Nguyen and Duc-Toan Nguyen}, doi = {10.1016/j.datak.2025.102509}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102509}, shortjournal = {Data Knowl. Eng.}, title = {A graph-based model for semantic textual similarity measurement}, volume = {161}, year = {2026}, }
• (2026). Rule-guided process discovery. DKE, 161, 102508. (www)

  @article{ , abstract = {Event data extracted from information systems serves as the foundation for process mining, enabling the extraction of insights and identification of improvements. Process discovery focuses on deriving descriptive process models from event logs, which form the basis for conformance checking, performance analysis, and other applications. Traditional process discovery techniques predominantly rely on event logs, often overlooking supplementary information such as domain knowledge and process rules. These rules, which define relationships between activities, can be obtained through automated techniques like declarative process discovery or provided by domain experts based on process specifications. When used as an additional input alongside event logs, such rules have significant potential to guide process discovery. However, leveraging rules to discover high-quality imperative process models, such as BPMN models and Petri nets, remains an underexplored area in the literature. To address this gap, we propose an enhanced framework, IMr, which integrates discovered or user-defined rules into the process discovery workflow via a novel recursive approach. The IMr framework employs a divide-and-conquer strategy, using rules to guide the selection of process structures at each recursion step in combination with the input event log. We evaluate our approach on several real-world event logs and demonstrate that the discovered models better align with the provided rules without compromising their conformance to the event log. Additionally, we show that high-quality rules can improve model quality across well-known conformance metrics. This work highlights the importance of integrating domain knowledge into process discovery, enhancing the quality, interpretability, and applicability of the resulting process models.}, archive = {J_DKE}, author = {Ali Norouzifar and Marcus Dees and Wil van der Aalst}, doi = {10.1016/j.datak.2025.102508}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102508}, shortjournal = {Data Knowl. Eng.}, title = {Rule-guided process discovery}, volume = {161}, year = {2026}, }
• (2026). LQ-FJS: A logical query-digging fake-news judgment system with structured video-summarization engine using LLM. DKE, 161, 102507. (www)

  @article{ , abstract = {The proliferation of online social platforms can greatly benefit people by fostering remote relationships, but it also inevitably amplifies the impact of multimodal fake news on societal trust and ethics. Existing fake-news detection AI systems are still vulnerable to the inconspicuous and indiscernible multimodal misinformation, and often lacking interpretability and accuracy in cross-platform settings. Hence, we propose a new innovative logical query-digging fake-news judgment system (LQ-FJS) to tackle the above problem based on multimodal approach. The LQ-FJS verifies the truthfulness of claims made within multimedia news by converting video content into structured textual summaries. It then acts as an interpretable agent, explaining the reasons for identified fake news by the structured video-summarization engine (SVSE) to act as an interpretable detection intermediary agent. The SVSE generates condensed captions for raw video content, converting it into structured textual narratives. Then, LQ-FJS exploits these condensed captions to retrieve reliable information related to the video content from LLM. Thus, LQ-FJS cross-verifies external knowledge sources and internal LLM responses to determine whether contradictions exist with factual information through a multimodal inconsistency verification procedure. Our experiments demonstrate that the subtle summarization produced by SVSE can facilitate the generation of explanatory reports that mitigate large-scale trust deficits caused by opaque “black-box” models. Our experiments show that LQ-FJS improves F1 scores by 4.5% and 7.2% compared to state-of-the-art models (FactLLaMA 2023 and HiSS 2023), and increases 14% user trusts through interpretable conclusions.}, archive = {J_DKE}, author = {Jhing-Fa Wang and Din-Yuen Chan and Hsin-Chun Tsai and Bo-Xuan Fang}, doi = {10.1016/j.datak.2025.102507}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102507}, shortjournal = {Data Knowl. Eng.}, title = {LQ-FJS: A logical query-digging fake-news judgment system with structured video-summarization engine using LLM}, volume = {161}, year = {2026}, }
• (2026). ABBA: Index structure for sequential pattern-based aggregate queries. DKE, 161, 102506. (www)

  @article{ , abstract = {Pattern-based aggregate (PBA) queries constitute an important and widely used type of analytical queries in sequence OLAP (S-OLAP) systems. Unfortunately, finding accurate answers to PBA queries in the S-OLAP system is often very expensive both in terms of time and memory consumption. In this paper we propose an efficient and easily maintainable index structure called the ABBA Index, which addresses the problem of PBA query processing. Experiments conducted using the KDD Cup data and public transport passengers’ travel behavior data show that our index outperforms state-of-the art solutions while requiring much less memory. The ABBA Index can be easily extended to support pattern-based aggregate queries over hierarchy (PBA-H), a novel class of analytical queries which we introduce as the second main contribution of the paper. Sensitivity, scalability and complexity analysis of the ABBA Index is also provided.}, archive = {J_DKE}, author = {Witold Andrzejewski and Tadeusz Morzy and Maciej Zakrzewicz}, doi = {10.1016/j.datak.2025.102506}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102506}, shortjournal = {Data Knowl. Eng.}, title = {ABBA: Index structure for sequential pattern-based aggregate queries}, volume = {161}, year = {2026}, }
• (2026). Elevating human-machine collaboration in NLP for enhanced content creation and decision support. DKE, 161, 102505. (www)

  @article{ , abstract = {Human-machine collaboration in Natural Language Processing (NLP) is revolutionizing content creation and decision support by seamlessly combining the strengths of both entities for enhanced efficiency and quality. The lack of seamless integration between human creativity and machine efficiency in NLP hinders optimal content creation and decision support. The objective of this study is to explore and promote the integration of human-machine collaboration in NLP to enhance both content creation and decision support processes. Data Acquisition for NLP requests involves defining the task and target audience, identifying relevant data sources like text documents and web data, and incorporating human expertise for data curation through validation and annotation. Machine processing techniques like tokenization, stemming/lemmatization, and removal of stop words, as well as human input for tasks like data annotation and error correction, to improve data quality and relevance for NLP applications. The combination of automated processing and human feedback leads to more precise and dependable effects. Techniques such as sentiment analysis, topic modelling, and entity recognition are utilized to excerpt valued perceptions from the data and enhance collaboration between humans and machines. These techniques help to streamline the NLP process and ensure that the system is providing accurate and relevant information to users. The analysis of NLP models in machine processing involves training the models to perform specific tasks, such as summarization, sentiment analysis, information extraction, trend identification, and creative content generation. The results show that social media leads with 90% usage, pivotal for audience engagement, while blogs at 78% highlight their depth in content creation implementation using Python software. These trained models are then used to improve decision-making processes, generate creative content, and enhance the accuracy of search results. The future scope involves leveraging advanced NLP techniques to deepen the collaboration between humans and machines for more effective content creation and decision support.}, archive = {J_DKE}, author = {Priyanka V. Deshmukh and Aniket K. Shahade}, doi = {10.1016/j.datak.2025.102505}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102505}, shortjournal = {Data Knowl. Eng.}, title = {Elevating human-machine collaboration in NLP for enhanced content creation and decision support}, volume = {161}, year = {2026}, }
• (2026). ELEVATE-ID: Extending large language models for end-to-end entity linking evaluation in indonesian. DKE, 161, 102504. (www)

  @article{ , abstract = {Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of natural language processing tasks. However, their effectiveness in low-resource languages remains underexplored, particularly in complex tasks such as end-to-end Entity Linking (EL), which requires both mention detection and disambiguation against a knowledge base (KB). In earlier work, we introduced IndEL — the first end-to-end EL benchmark dataset for the Indonesian language — covering both a general domain (news) and a specific domain (religious text from the Indonesian translation of the Quran), and evaluated four traditional end-to-end EL systems on this dataset. In this study, we propose ELEVATE-ID, a comprehensive evaluation framework for assessing LLM performance on end-to-end EL in Indonesian. The framework evaluates LLMs under both zero-shot and fine-tuned conditions, using multilingual and Indonesian monolingual models, with Wikidata as the target KB. Our experiments include performance benchmarking, generalization analysis across domains, and systematic error analysis. Results show that GPT-4 and GPT-3.5 achieve the highest accuracy in zero-shot and fine-tuned settings, respectively. However, even fine-tuned GPT-3.5 underperforms compared to DBpedia Spotlight — the weakest of the traditional model baselines — in the general domain. Interestingly, GPT-3.5 outperforms Babelfy in the specific domain. Generalization analysis indicates that fine-tuned GPT-3.5 adapts more effectively to cross-domain and mixed-domain scenarios. Error analysis uncovers persistent challenges that hinder LLM performance: difficulties with non-complete mentions, acronym disambiguation, and full-name recognition in formal contexts. These issues point to limitations in mention boundary detection and contextual grounding. Indonesian-pretrained LLMs, Komodo and Merak, reveal core weaknesses: template leakage and entity hallucination, respectively—underscoring architectural and training limitations in low-resource end-to-end EL. 1}, archive = {J_DKE}, author = {Ria Hari Gusmita and Asep Fajar Firmansyah and Hamada M. Zahera and Axel-Cyrille Ngonga Ngomo}, doi = {10.1016/j.datak.2025.102504}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102504}, shortjournal = {Data Knowl. Eng.}, title = {ELEVATE-ID: Extending large language models for end-to-end entity linking evaluation in indonesian}, volume = {161}, year = {2026}, }
• (2026). Time-aware complex question answering over temporal knowledge graph. DKE, 161, 102503. (www)

  @article{ , abstract = {Knowledge Graph Question Answering (KGQA) is a crucial topic in Knowledge Graphs (KGs), with the objective of retrieving the corresponding facts from KGs to answer given questions. In practical applications, facts in KGs usually have time constraints, thus, question answering on Temporal Knowledge Graphs (TKGs) has attracted extensive attention. Existing Temporal Knowledge Graph Question Answering (TKGQA) methods focus on dealing with complex questions involving multiple facts, and mainly face two challenges. First, these methods only consider matching questions with facts in TKGs to identify the answer, ignoring the temporal order between different facts, which makes it challenging to solve the questions involving temporal order. Second, they usually focus on the representation of the question text while neglecting the rich semantic information within the questions, which leads to certain limitations in understanding question. To address the above challenges, this research proposes a model named Time-Aware Complex Question Answering (TA-CQA). Specifically, we extend the Temporal Knowledge Graph Embedding (TKGE) model by incorporating temporal order information into the embedding vectors, ensuring that the model can distinguish the temporal order of different facts. To enhance the semantic representation of the question, we integrate question information using attention mechanism and learnable encoder. Different from the previous TKGQA methods, we propose time relevance measurement to further enhance the accuracy of answer prediction by better capturing the correlation between question information and time information. Multiple sets of experiments on CronQuestions and TimeQuestions demonstrate our model’s superior performance across all question types. In particular, for complex questions involving multiple facts, the hit@1 values are increased by 3.2% and 3.5% respectively.}, archive = {J_DKE}, author = {Luyi Bai and Tongyue Zhang and Guangchen Feng}, doi = {10.1016/j.datak.2025.102503}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102503}, shortjournal = {Data Knowl. Eng.}, title = {Time-aware complex question answering over temporal knowledge graph}, volume = {161}, year = {2026}, }
• (2026). Conceptual modeling of user perspectives — From data warehouses to alliance-driven data ecosystems. DKE, 161, 102502. (www)

  @article{ , abstract = {The increasing complexity of modern information systems has highlighted the need for advanced conceptual modeling techniques that incorporate multi-perspective and view-based approaches. This paper explores the role of multi-perspective modeling and view modeling in designing distributed, heterogeneous systems while addressing diverse user requirements and ensuring semantic consistency. These methods enable the representation of multiple viewpoints, traceability, and dynamic integration across different levels of abstraction. Key advancements in schema mapping, view maintenance, and semantic metadata management are examined, illustrating how they support query optimization, data quality, and interoperability. We discuss how data management architectures, such as data ecosystems, data warehouses, and data lakes, leverage these innovations to enable flexible and sustainable data sharing. By integrating user-centric and goal-oriented modeling frameworks, the alignment of technical design with organizational and social requirements is emphasized. Future challenges include the need for enhanced reasoning capabilities and collaborative tools to manage the growing complexity of interconnected systems while maintaining adaptability and trust.}, archive = {J_DKE}, author = {Sandra Geisler and Christoph Quix and István Koren and Matthias Jarke}, doi = {10.1016/j.datak.2025.102502}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102502}, shortjournal = {Data Knowl. Eng.}, title = {Conceptual modeling of user perspectives — From data warehouses to alliance-driven data ecosystems}, volume = {161}, year = {2026}, }
• (2026). Source-free domain adaptation with complex distribution considerations for time series data. DKE, 161, 102501. (www)

  @article{ , abstract = {Source-Free Domain Adaptation (SFDA) aims to adapt a pre-trained model from a labeled source domain to an unlabeled target domain without accessing source domain data, thereby protecting source domain privacy. Although SFDA has recently been applied to time series data, the inherent complex distribution characteristics including temporal variability and distributional diversity of such data remain underexplored. Time series data exhibit significant dynamic variability influenced by collection environments, leading to discrepancies between sequences. Additionally, multidimensional time series data face distributional diversity across dimensions. These complex characteristics increase the learning difficulty for source models and widen the adaptation gap between the source and target domains. To address these challenges, this paper proposes a novel SFDA method for time series data, named Adaptive Latent Subdomain feature extraction and joint Prediction (ALSP). The method divides the source domain, which has a complex distribution, into multiple latent subdomains with relatively simple distributions, thereby effectively capturing the features of different subdistributions. It extracts latent domain-specific and domain-invariant features to identify subdomain-specific characteristics. Furthermore, it combines domain-specific classifiers and a domain-invariant classifier to enhance model performance through multi-classifier joint prediction. During target domain adaptation, ALSP reduces domain dependence by extracting invariant features, thereby narrowing the distributional gap between the source and target domains. Simultaneously, it leverages prior knowledge from the source domain distribution to support the hypothesis space and dynamically adapt to the target domain. Experiments on three real-world datasets demonstrate that ALSP achieves superior performance in cross-domain time series classification tasks, significantly outperforming existing methods.}, archive = {J_DKE}, author = {Jing Shang and Zunming Chen and Zhiwen Xiao and Zhihui Wu and Yifei Zhang and Jibing Wang}, doi = {10.1016/j.datak.2025.102501}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102501}, shortjournal = {Data Knowl. Eng.}, title = {Source-free domain adaptation with complex distribution considerations for time series data}, volume = {161}, year = {2026}, }
• (2026). Conceptual modeling: A large language model assistant for characterizing research contributions. DKE, 161, 102497. (www)

  @article{ , abstract = {The body of conceptual modeling research publications is vast and diverse, making it challenging for a single researcher or research group to fully comprehend the field’s overall development. Although some approaches have been proposed to help organize these research contributions, it is still unrealistic to expect human experts to manually comprehend and characterize all of this research. However, as generative AI tools based on large language models, such as ChatGPT, become increasingly sophisticated, it may be possible to replace or augment tedious, manual work with semi-automated approaches. In this research, we present a customized version of ChatGPT that is tuned to the task of characterizing conceptual modeling research. Experiments with this AI tool demonstrate that it is feasible to create a usable knowledge survey for the continually evolving body of conceptual modeling research contributions.}, archive = {J_DKE}, author = {Stephen W. Liddle and Heinrich C. Mayr and Oscar Pastor and Veda C. Storey and Bernhard Thalheim}, doi = {10.1016/j.datak.2025.102497}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102497}, shortjournal = {Data Knowl. Eng.}, title = {Conceptual modeling: A large language model assistant for characterizing research contributions}, volume = {161}, year = {2026}, }
• (2026). Dynamic time warping for classifying long-term trends in time series. DKE, 161, 102495. (www)

  @article{ , abstract = {This paper explores the potential of dynamic time warping (DTW) for recognizing different segments in time series data characterized by their long-term trends and curvature. To perform classification, a set of reference data for each class is required, where each time series in the reference set represents a typical shape of that class. The classification process involves computing the DTW distance between a given time series and each reference time series, then assigning the time series to the class with the minimum distance. Experiments on both simulated and real-world time series data from two different use cases demonstrate that DTW can correctly classify the different segments. Additionally, the paper investigates whether incorrectly classified phases could indicate data security issues. Additional experiments are performed to assess the number of data points required to reliably classify a segment correctly. These experiments highlight the limitations and emphasize the importance of selecting good reference data.}, archive = {J_DKE}, author = {Anna-Christina Glock and Klaus Chmelina and Johannes Fürnkranz and Thomas Hütter}, doi = {10.1016/j.datak.2025.102495}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102495}, shortjournal = {Data Knowl. Eng.}, title = {Dynamic time warping for classifying long-term trends in time series}, volume = {161}, year = {2026}, }
• (2026). Semantic-aware query answering with large language models. DKE, 161, 102494. (www)

  @article{ , abstract = {In the modern data-driven world, answering queries over heterogeneous and semantically inconsistent data remains a significant challenge. Modern datasets originate from diverse sources, such as relational databases, semi-structured repositories, and unstructured documents, leading to substantial variability in schemas, terminologies, and data formats. Traditional systems, constrained by rigid syntactic matching and strict data binding, struggle to capture critical semantic connections and schema ambiguities, failing to meet the growing demand among data scientists for advanced forms of flexibility and context-awareness in query answering. In parallel, the advent of Large Language Models (LLMs) has introduced new capabilities in natural language interpretation, making them highly promising for addressing such challenges. However, LLMs alone lack the systematic rigor and explainability required for robust query processing and decision-making in high-stakes domains. In this paper, we propose Soft Query Answering (Soft QA), a novel hybrid approach that integrates LLMs as an intermediate semantic layer within the query processing pipeline. Soft QA enhances query answering adaptability and flexibility by injecting semantic understanding through context-aware, schema-informed prompts, and leverages LLMs to semantically link entities, resolve ambiguities, and deliver accurate query results in complex settings. We demonstrate its practical effectiveness through real-world examples, highlighting its ability to resolve semantic mismatches and improve query outcomes without requiring extensive data cleaning or restructuring.}, archive = {J_DKE}, author = {Paolo Atzeni and Teodoro Baldazzi and Luigi Bellomarini and Eleonora Laurenza and Emanuel Sallinger}, doi = {10.1016/j.datak.2025.102494}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102494}, shortjournal = {Data Knowl. Eng.}, title = {Semantic-aware query answering with large language models}, volume = {161}, year = {2026}, }
• (2026). An integrated requirements framework for analytical and AI projects. DKE, 161, 102493. (www)

  @article{ , abstract = {To this day, the requirements of data warehouses, user visualizations and ML projects have been tackled in an independent manner, ignoring the possible cross-requirements, collective constraints and dependencies between the outputs of the different systems that should be taken into account to ensure a successful analytical project. In this work, we take a holistic approach and propose a methodology that supports modeling and subsequent analysis while taking into account these three aspects. This methodology has several advantages, mainly that (i) it enables us to identify possible conflicts between actors on different tasks that are overlooked if the systems are treated in an isolated manner and (ii) this holistic view enables modeling multi-company systems, where the information or even the analytical results can be provided by third-parties, identifying key participants in federated environments. After presenting the required formalism to carry out this kind of analysis, we showcase it on a real-world running example of the tourism sector.}, archive = {J_DKE}, author = {Juan Trujillo and Ana Lavalle and Alejandro Reina-Reina and Jorge García-Carrasco and Alejandro Maté and Wolfgang Maaß}, doi = {10.1016/j.datak.2025.102493}, journal = {Data & Knowledge Engineering}, month = {1}, pages = {102493}, shortjournal = {Data Knowl. Eng.}, title = {An integrated requirements framework for analytical and AI projects}, volume = {161}, year = {2026}, }

DSS - 9

• (2025). Corporate credit scoring method based on unlabeled data and multi-source data. DSS, 198, 114543. (www)

  @article{ , abstract = {Unlabeled data and multi-source data provide unprecedented opportunities for the financial industry to improve credit scoring accuracy. When utilizing unlabeled data, existing credit scoring methods often suffer from unreliability issues due to improper clustering or the introduction of noise when predicting labels. When utilizing multi-source data, existing credit scoring methods based on federated learning frameworks fail to tailor models for different data distributions of different data sources due to the limitations of relying on a single global model. Moreover, recent studies have explored the individual value of unlabeled data and multi-source data, but they often fail to utilize both. To address these issues, we propose UMDCS (Unlabeled and Multi-Source data Driven Credit Scoring), a self-supervised credit scoring method that utilizes both unlabeled and multi-source data simultaneously. To utilize unlabeled data, we propose a novel sample masking function to generate pseudo-labels for unlabeled data and pre-train the encoder using the pretext tasks. To utilize multi-source data, we employ a horizontal federated learning framework to aggregate local encoders into a global model while preserving data privacy. The global encoder is concatenated with personalized predictors to form personalized credit scoring models for each data source. Five experiments and statistical significance tests show that UMDCS outperforms other baseline methods.}, archive = {J_DSS}, author = {Yunhong Xu and Yitong Chen and Li Sun and Yu Chen}, doi = {10.1016/j.dss.2025.114543}, journal = {Decision Support Systems}, month = {11}, pages = {114543}, shortjournal = {Decis. Supp. Syst.}, title = {Corporate credit scoring method based on unlabeled data and multi-source data}, volume = {198}, year = {2025}, }
• (2025). Being responsible or affable: Investigating the effects of AI error correction behaviors on user engagement. DSS, 198, 114542. (www)

  @article{ , abstract = {Affable design is increasingly employed in AI conversational agents to foster smoother interaction and enhance user experience. However, a growing concern is that this overemphasis on social appeal often overlooks corrective interventions, particularly when users hold false or biased beliefs. Such omissions carry the risk of reinforcing user misconceptions and ultimately undermining the effectiveness of human–AI collaboration. Drawing upon the attribution theory, this study investigates whether the error-correction behavior of AI agents offset these risks and improve user engagement. Empirical evidence from three experimental studies verifies that AI agents' error-correction behavior indeed enhances users' perceived responsibility of AI agents and strengthens their engagement intentions. This effect does not appear to compromise social comfort, especially in the context where responsibility takes precedence, such as healthcare. This study further finds that the high expertise of AI agents amplifies the positive effects of error-correction behavior, while high entitativity diminishes these effects by blurring AI agents' responsibility. These findings offer important guidance for designing responsible AI agents and highlight the value of AI error-correction behaviors in human-AI interaction.}, archive = {J_DSS}, author = {Yunchang Zhu and Xianghua Lu}, doi = {10.1016/j.dss.2025.114542}, journal = {Decision Support Systems}, month = {11}, pages = {114542}, shortjournal = {Decis. Supp. Syst.}, title = {Being responsible or affable: Investigating the effects of AI error correction behaviors on user engagement}, volume = {198}, year = {2025}, }
• (2025). Emotion aware session based news recommender systems. DSS, 198, 114540. (www)

  @article{ , abstract = {News recommender systems are decision support systems that exploit user-article interactions over a short duration of time to discover users’ interests and predict unseen news articles to generate a ranking of news articles that are relevant and interesting. In the news recommendation scenario, the relevance of articles decays quickly, and fresh articles are generated daily. Session based models are proposed using time-aware approaches to exploit interactions sequentially. Prior news recommender systems do not consider emotional information expressed in news articles within sessions for recommendations. Emotions play a key role in supporting decision-making and emotionally charged headlines can evoke curiosity or urgency, prompting users to click on certain articles. This paper presents an innovative decision support system for session based news recommendation, using expressed emotions from news articles, such as expressed in the title, abstract, and text, to improve user decision-making. We introduce a novel methodology that incorporates expressed emotions into three session based news recommendation models. Our results demonstrate that expressed emotion carries valuable information to improve session based news recommenders on various ranking metrics significantly and proved especially beneficial in scenarios with limited user interaction history, addressing the cold-start problem. The results show significant improvements in ranking metrics, emphasizing the utility of emotional features for dynamic decision-making support.}, archive = {J_DSS}, author = {Benjamin Gundersen and Saikishore Kalloori and Abhishek Srivastava}, doi = {10.1016/j.dss.2025.114540}, journal = {Decision Support Systems}, month = {11}, pages = {114540}, shortjournal = {Decis. Supp. Syst.}, title = {Emotion aware session based news recommender systems}, volume = {198}, year = {2025}, }
• (2025). Decision support for integrated trade agent's procurement and sales planning under uncertainty. DSS, 198, 114537. (www)

  @article{ , abstract = {This paper investigates a trade agent decision optimization problem (TADOP), in which a trade agent (TA) selects a subset of retailers and suppliers to maximize its profit under uncertain demand and spot price. The TA operates between suppliers and retailers as a third-party platform and decide which subset of retailers to serve, taking into account capacity reservations with option suppliers in advance. Once demand and spot price are realized, the TA decides how much to procure from each channel to fulfill retailers' demand. The problem is formulated as a two-stage stochastic program. Due to the high complexity and large number of scenarios, we reformulate the problem as a set-partition model, where the master problem (MP) selects the combination of retailers to serve, and the subproblem (SP) identifies the optimal procurement plans, thus reducing the number of variables and constraints. To further enhance tractability, the SP is transformed into an equivalent shortest-path problem (SPP) to address issues of non-linearity and non-convexity. Experimental results demonstrate the effectiveness of the decomposition approach, providing TAs with a practical decision-making tool for procurement and sales. Furthermore, the insights gained into TAs' procurement and sales strategies across various scenarios offer valuable guidance for decision-making in uncertain supply chain environments.}, archive = {J_DSS}, author = {An Liu and Xinyu Wang and Jiafu Tang}, doi = {10.1016/j.dss.2025.114537}, journal = {Decision Support Systems}, month = {11}, pages = {114537}, shortjournal = {Decis. Supp. Syst.}, title = {Decision support for integrated trade agent's procurement and sales planning under uncertainty}, volume = {198}, year = {2025}, }
• (2025). Modeling hybrid firm relationships with graph neural networks for stock investment decisions. DSS, 198, 114528. (www)

  @article{ , abstract = {The highly volatile nature of the stock market makes predicting data patterns challenging. Significant efforts have been dedicated to modeling complex stock correlations to improve stock return forecasting and support better investor decision-making. Although various predefined intrinsic associations and learned implicit graph structures have been discovered, they have limitations in fully exploring and leveraging both types of graph information. In this paper, we proposed a Hybrid Structure-aware Graph Neural Network (HSGNN) framework. Unlike models that rely solely on predefined or learned graphs, HSGNN utilizes money-flow graphs to complementarily learn implicit graph structures and applies sparse supply-chain graphs to jointly enhance stock return forecasting. Extensive experiments on real stock benchmarks demonstrate our proposed HSGNN outperforms various state-of-the-art forecasting methods, offering a robust decision-support system for financial stakeholders.}, archive = {J_DSS}, author = {Yang Du and Biao Li and Zhichen Lu and Gang Kou}, doi = {10.1016/j.dss.2025.114528}, journal = {Decision Support Systems}, month = {11}, pages = {114528}, shortjournal = {Decis. Supp. Syst.}, title = {Modeling hybrid firm relationships with graph neural networks for stock investment decisions}, volume = {198}, year = {2025}, }
• (2025). Social capital matters: Towards comprehensive user preference for product recommendation with deep learning. DSS, 198, 114527. (www)

  @article{ , abstract = {Social recommender systems help address data sparsity in user–product interactions by leveraging social relationships to infer user preferences. However, existing models often overlook the role of social capital that influence decision-making in social commerce. Social capital consists of structural, relational, and cognitive dimensions, all of which shape user preferences. To better understand these influences, we propose a multi-task learning framework named DeepSC that integrates social capital theory into preference modeling. Its user preference learning module extracts structural features through graph-based pre-training, learns relational features from dynamic user embeddings, and models cognitive features using a hypergraph attention network. Additionally, the dual graph-based product feature learning module enhances cognitive feature extraction by incorporating product co-interactions. DeepSC is optimized through a joint learning objective, combining point-wise and pair-wise learning with an auxiliary social link prediction task to refine user representations. Experiments on three e-commerce datasets demonstrate that DeepSC significantly outperforms the state-of-the-art recommendation models, highlighting the effectiveness of integrating social capital into social preference learning. Our research advances social recommendation by providing a social capital theory-driven approach to modeling user behavior in digital commerce.}, archive = {J_DSS}, author = {Weiyue Li and Ming Gao and Bowei Chen and Jingmin An and Yeming Gong}, doi = {10.1016/j.dss.2025.114527}, journal = {Decision Support Systems}, month = {11}, pages = {114527}, shortjournal = {Decis. Supp. Syst.}, title = {Social capital matters: Towards comprehensive user preference for product recommendation with deep learning}, volume = {198}, year = {2025}, }
• (2025). Enhancing cybersecurity risk assessment using temporal knowledge graph-based explainable decision support system. DSS, 198, 114526. (www)

  @article{ , abstract = {Assessing cybersecurity policies is crucial for any organization to combat evolving cyber threats. The absence of a comprehensive dataset has prevented previous studies from analyzing the risk of organizations’ cybersecurity policies. Past studies have not considered temporal information in the policies. Analysis of cybersecurity policies using attention mechanism requires automated determination of optimal number of attention units which remains unaddressed. Moreover, absence of interpretation in cybersecurity studies creates a barrier to understanding policy vulnerabilities and developing targeted solutions. To address these challenges, we develop a decision support system which (i) enhances risk classification of organization’s cybersecurity policies, (ii) develops a comprehensive cybersecurity policy dataset from the websites of 190 companies, transformed into a knowledge graph to capture entity relationships among various policies, (iii) integrates temporal information into the knowledge graph by incorporating time stamps from event sequences in cyberattack information, (iv) develops Explainable Factor Analysis based Multi-Head Attention mechanism, which automates the determination of the optimal number of attention units and optimizes data allocation across attention units using factor analysis, and (v) utilizes attention heatmaps and shapley values for interpretability. Our cybersecurity policy dataset is used as a case study with four benchmark datasets for further validation. Results reveal that our model outperforms the other state-of-the-art, achieving an 87.78% F 1 score, followed by robustness checking and statistical significance testing. Finally, Shapley values are used to interpret the model’s output to identify vulnerabilities within the organizational policies, providing crucial insights enabling decision-makers to enhance their cybersecurity policies and mitigate potential threats.}, archive = {J_DSS}, author = {Subhajit Bag and Sobhan Sarkar and Indranil Bose}, doi = {10.1016/j.dss.2025.114526}, journal = {Decision Support Systems}, month = {11}, pages = {114526}, shortjournal = {Decis. Supp. Syst.}, title = {Enhancing cybersecurity risk assessment using temporal knowledge graph-based explainable decision support system}, volume = {198}, year = {2025}, }
• (2025). A field study on the impact of the counter ad-blocking wall strategy on user engagement. DSS, 198, 114525. (www)

  @article{ , abstract = {Ad-blocking tools prevent ads from being shown to web users. Their increasingly widespread usage poses an existential risk to online publishers who provide free content and rely on display ads for revenue. Studies on counter ad-blocking strategies taken by publishers are limited, especially with regard to how these strategies affect user engagement, thus posing additional uncertainties to the selection of a suitable counter ad-blocking strategy. Through a randomized field experiment with a large global publisher, our study seeks to understand how the two most common counter ad-blocking strategies, (i) Wall and (ii) Acceptable Ads Exchange (AAX), affect user engagement differently. Our results show that the Wall strategy causes a lower overall engagement compared to AAX, mainly due to users who refuse to whitelist and leave the website. Over time, the negative impact increases, albeit at a slower speed. Furthermore, heavier users, identified based on the amount of engagement in the pre-treatment period, are less affected by the Wall strategy than lighter users; instrumental users, who read for practical purposes, are less affected than entertainment users. Finally, the Wall strategy has a bigger negative impact on the engagement of popular and new articles, compared to niche and old articles, respectively, as observed by a longer tail in engagement distribution with respect to content. These results on the heterogeneous effects of counter ad-blocking strategies on engagement offer novel and important managerial implications on a publisher’s choice of counter ad-blocking strategy and editorial decisions.}, archive = {J_DSS}, author = {Michael K. Chen and Shuai Zhao and Cristian Borcea and Yi Chen}, doi = {10.1016/j.dss.2025.114525}, journal = {Decision Support Systems}, month = {11}, pages = {114525}, shortjournal = {Decis. Supp. Syst.}, title = {A field study on the impact of the counter ad-blocking wall strategy on user engagement}, volume = {198}, year = {2025}, }
• (2025). A cross-platform rumor detection framework considering data privacy protection and different detection capabilities of online social platforms. DSS, 198, 114524. (www)

  @article{ , abstract = {The anonymity and widespread popularity of online social platforms (OSPs) allow users to share uncertain posts freely, leading to numerous rumors. Similar rumors spread widely across OSPs, resulting in frequent cross-platform rumors (CPRs). Owing to the unique nature of the cross-platform spread, the dual challenges of data privacy protection constraints and differences in the data and detection capabilities of OSPs exacerbate the difficulty of CPR detection. Thus, to detect CPRs effectively, we designed and implemented a novel deep learning framework named Cross Platform Rumor Detection based on Improved Federated Learning (CPRDIFL), which integrates and improves federated learning and the pre-trained Masked and Contextualized BERT (MacBERT). Our framework uses FL to analyze data from OSPs independently, thus avoiding the need for data integration and ensuring the data privacy protection of OSPs. Moreover, MacBERT is deployed on the clients of CPRDIFL to extract contextual features from posts and dynamically update local weights based on the data and detection performance. Weight parameters are dynamically shared between clients and servers and between clients to achieve complementary advantages across OSPs. Our framework was used in six comprehensive experiments in different scenarios, and the experimental results showed that it achieved the best results in CPR detection. This study not only provides an effective solution for CPR detection but also marks a significant step toward the automated detection of cross-OSP information pollution.}, archive = {J_DSS}, author = {Xuelong Chen and Jinchao Pan}, doi = {10.1016/j.dss.2025.114524}, journal = {Decision Support Systems}, month = {11}, pages = {114524}, shortjournal = {Decis. Supp. Syst.}, title = {A cross-platform rumor detection framework considering data privacy protection and different detection capabilities of online social platforms}, volume = {198}, year = {2025}, }

EAAI - 348

• (2025). A multi-scale feature extraction and fusion framework based on wavelet Kolmogorov–Arnold networks and parallel bi-directional gated recurrent units for electric load forecasting. EAAI, 162, 112517. (www)

  @article{ , abstract = {Short-term electric load forecasting remains challenged by the dual requirements of accuracy and robustness due to the combined effects of strong seasonality, multi-scale spikes, and stochastic disturbances. To address this, we propose a novel multi-scale forecasting framework, NP-WavKAN-Fusion, which integrates Neural Prophet for data decomposition and a Wavelet-based Kolmogorov–Arnold Network (WavKAN) with learnable wavelet kernels for multi-scale encoding. This fusion model utilizes a Bi-directional Gated Recurrent Unit (BiGRU) to capture long-term temporal dependencies and an adaptive feature fusion gate (AFF) to dynamically re-weight static and dynamic features for final load predictions. Extensive experiments on two public datasets from Australia and Morocco show that NP-WavKAN-Fusion consistently outperforms traditional models, reducing the mean absolute error by at least 30 %. For multi-step forecasting tasks, NP-WavKAN-Fusion maintains error inflation within 15 %, demonstrating superior performance compared to state-of-the-art long-sequence models such as Informer and PatchTST. The Diebold–Mariano test confirms that NP-WavKAN-Fusion yields statistically significant improvements, with 19 out of 20 comparisons showing lower errors. Ablation studies show that removing either the Neural Prophet component or the AFF significantly increases the forecasting error, validating the necessity of our layered denoising and fusion strategies. The proposed NP-WavKAN-Fusion framework demonstrates strong potential for real-world applications in electric load forecasting, offering robust performance under various temporal and non-stationary conditions.}, archive = {J_EAAI}, author = {Chunliang Mai and Lixin Zhang and Xuewei Chao and Xue Hu and Omar Behar}, doi = {10.1016/j.engappai.2025.112517}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112517}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale feature extraction and fusion framework based on wavelet Kolmogorov–Arnold networks and parallel bi-directional gated recurrent units for electric load forecasting}, volume = {162}, year = {2025}, }
• (2025). A constitutive framework for non-isothermal plasticity through micro-mechanism informed artificial neural networks. EAAI, 162, 112465. (www)

  @article{ , abstract = {Non-isothermal plastic deformation creates modeling challenges for constitutive model development. Current constitutive models challenge to capture the coupled microstructural changes that occur when temperature varies during processing. The Micro-Mechanism Informed Artificial Neural Network (MMIANN) framework was developed to address these limitations using artificial intelligence (AI) and machine learning (ML) techniques. The MMIANN framework combines physics-based evolution equations for key metallurgical processes—dislocation density, grain boundary migration, and precipitation kinetics—with neural network predictions. These micro-mechanisms operate as internal state variables that guide the network's material behavior predictions. The architecture uses parallel physics-based and neural pathways, blended through adaptive coefficients. Thermodynamic constraints maintain consistency through penalty-based enforcement of the Clausius-Duhem inequality. The model was trained and validated using experimental data from simultaneous cooling and tensile deformation of AA7075 aluminum alloy. The tests replicated industrial hot forming conditions with cooling rates from 25 to 75 °C per second (°C/s). This experimental approach captures the thermal-mechanical coupling that drives microstructural evolution in practice. MMIANN achieved correlation coefficients exceeding 0.96 for stress, temperature, and grain size predictions. The framework captures thermomechanical regimes. Processing maps generated by the AI model link process parameters to microstructural outcomes. The analysis reveals an optimal processing window (40–55 °C/s cooling, 0.15–0.25 strain ( ε )) and identifies three regimes where different strengthening mechanisms dominate. By integrating metallurgical science with machine learning, this framework provides a practical tool for non-isothermal manufacturing processes. The approach bridges microstructural understanding with process control for thermal-mechanical operations through the application of artificial intelligence.}, archive = {J_EAAI}, author = {Yo-Lun Yang and Tsai-Fu Chung and Chia-Hung Liao and Liang-Yu Chen and Hsing-Yu Wu and Uthayakumar Marimuthu and Arumugaprabu Veerasimman and Sundarakannan Rajendran and Vigneshwaran Shanmugam}, doi = {10.1016/j.engappai.2025.112465}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112465}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A constitutive framework for non-isothermal plasticity through micro-mechanism informed artificial neural networks}, volume = {162}, year = {2025}, }
• (2025). A novel condition monitoring approach using hybrid lightweighted adaptive models for complex machinery. EAAI, 162, 112461. (www)

  @article{ , abstract = {Condition monitoring of complex industrial systems is critical for ensuring operational reliability. Data-driven methods using artificial intelligence have advanced anomaly detection (AD) and fault diagnosis (FD), but existing approaches often treat them separately, focus on known faults, and struggle with previously unseen or rare conditions in multi-modal scenarios. This study proposes a novel condition monitoring framework that integrates AD and FD within a distributed architecture. Lightweight models—including kernel principal component analysis, support vector machines, and one-dimensional convolutional neural networks—enable efficient and scalable processing. A multilevel information fusion strategy ensures consistent detection and diagnosis while facilitating the isolation of previously unknown faults. Module test results demonstrate the effectiveness and robustness of the proposed feature extraction and adaptive modeling approaches. The overall test results for previously unknown faults vary across channels and modules. For samples with misalignment and inner blade wear, channel-level detection accuracy ranges from 0.007 to 0.989, with unknown recognition rates up to 0.933 and diagnosis probabilities from 0.508 to 0.933. For strong misalignment and fan-end inner race faults, nearly all channels achieve 100 % detection accuracy, with some diagnosis probabilities above 0.9, while unknown recognition remains minimal (mostly below 0.05). Importantly, the proposed framework integrates detection and diagnostic outputs across channels, effectively mapping previously unseen faults to similar known categories or to an unknown category. Overall, the proposed framework offers a referenced solution for condition monitoring of industrial systems like pumps, turbines, and compressors, and lays the foundation for future improvements incorporating domain knowledge and model-driven interpretability.}, archive = {J_EAAI}, author = {Yingqian Liu and Rongyong Zhang and Luigi Grossi and Zhipin Ye and Huairui Li and Rongsheng Zhu and Qiang Fu}, doi = {10.1016/j.engappai.2025.112461}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112461}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel condition monitoring approach using hybrid lightweighted adaptive models for complex machinery}, volume = {162}, year = {2025}, }
• (2025). Finite element-integrated neural network framework for spatial modal prediction in machine tool structures. EAAI, 162, 112456. (www)

  @article{ , abstract = {Accurate prediction of position-dependent structural dynamics in (CNC) machine tools is critical for ensuring machining precision, avoiding resonance, and enabling real-time health monitoring. This study proposes an integrated framework combining finite element analysis (FEA), experimental modal testing, and artificial neural networks (ANNs) to model and predict position-dependent dynamics across the machine workspace. A validated FEA model of a high-rigidity vertical machining centre is developed and correlated with experimental modal analysis using a PCB 086D20 impact hammer and tri-axial accelerometer. Natural frequencies and mode shapes are extracted and compared, showing deviation under 10 %, confirming model fidelity. To capture position-dependent dynamics, modal analysis was performed at 27 spatial locations, revealing significant frequency variation across planes, indicating localized compliance zones. A multilayer ANN is trained on the modal dataset to predict frequencies based on spatial coordinates, achieving R 2 values above 0.99. The proposed hybrid approach enables real-time estimation of structural dynamics, reducing the need for repeated testing and supporting intelligent control strategies in large-format CNC systems. This work contributes a predictive foundation for dynamic stability optimization, resonance avoidance, and digital twin development in precision machining applications.}, archive = {J_EAAI}, author = {Aman Ullah and Tzu-Chi Chan and Jun-Fa Huang and Shinn-Liang Chang}, doi = {10.1016/j.engappai.2025.112456}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112456}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finite element-integrated neural network framework for spatial modal prediction in machine tool structures}, volume = {162}, year = {2025}, }
• (2025). Predicting thermal transport of blood-based penta-hybrid nanofluid in fin geometries using deep neural networks and finite difference approach. EAAI, 162, 112450. (www)

  @article{ , abstract = {Nanofluids have garnered significant research interest due to their enhanced heat transfer and thermal characteristics. A novel hybrid nanofluid has exhibited exceptional thermal properties, combining five nanoparticles of uniform shapes with a base fluid, such as blood. This study investigates the influence of fin thickness, varying with length, considering the implications of internal heat production, convection, and thermal radiation processes in rectangular, convex, and triangular fin descriptions. Wet scenarios are interpreted to evaluate differences in thermal energy dynamics for fin shapes like Rectangular, Convex and Triangular. Darcy's model is employed to account for the material's porous nature. A finite difference scheme, implemented using Partial Differential Equation solver (PDSolve) in Maple (2024), provides graphical insights into fin effectiveness and thermal steady-state responses across various parameters. Incorporating Penta hybrid nanofluids enhances fin performance, with rectangular fins' Nusselt numbers (up to 1.936) proving more efficient, delivering faster thermal responses than triangular fins and convex fins. Further, using the Adam Optimisation algorithm, Convolutional Neural Networks were used to validate the current model. It was observed that these networks could accurately forecast the truth values, and the two findings matched, as indicated in Table 3 As a potential biological application, this research offers insight into optimising cooling systems for biomedical devices, such as heat exchangers in artificial organs.}, archive = {J_EAAI}, author = {Maddina Dinesh Kumar and Nehad Ali Shah and Dharmaiah Gurram and Se-Jin Yook}, doi = {10.1016/j.engappai.2025.112450}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112450}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting thermal transport of blood-based penta-hybrid nanofluid in fin geometries using deep neural networks and finite difference approach}, volume = {162}, year = {2025}, }
• (2025). Enhanced multi-modal emotion recognition using the feature level fusion. EAAI, 162, 112447. (www)

  @article{ , abstract = {Multi-modal human emotion recognition is a complex process of synthesizing information from various modalities to calculate emotion states. This field faces several challenges: (1) Acoustic is an essential component of emotion expression, but it often underperforms compared to visual and text in emotion recognition. (2) Capturing the feature interaction among different modalities is usually complex. (3) Processing high-definition videos can significantly reduce the efficiency of visual analysis. In this study, we presented a learning architecture designed to recognize human emotions effectively. For the first challenge, we implemented a multi-level acoustic encoder (MLAE) that enhances the extraction of acoustic information to improve the acoustic contribution in multi-modal emotion recognition. Facing the second challenge, we introduced the cross-attention block module, which adeptly captures the inter-modal interactions. To address the third challenge, we adopted the re-parameterized visual geometry group network (RepVGG) as the visual feature encoder, employing its multi-branch learning and single-branch reasoning structure to maintain high reasoning efficiency. Our model has demonstrated the state-of-the-art performance of the interactive emotional dyadic motion capture (IEMOCAP) dataset and the multi-modal opinion sentiment and emotion intensity of the Carnegie Mellon University (CMU-MOSEI) dataset.}, archive = {J_EAAI}, author = {Aziguli Wulamu and Yuheng Wu and Xin Liu and Yao Zhang and Jinghan Xu and Yang Zhang}, doi = {10.1016/j.engappai.2025.112447}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112447}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced multi-modal emotion recognition using the feature level fusion}, volume = {162}, year = {2025}, }
• (2025). Deep adaptive wavelet autoencoder with mutually independent empirical cumulative distribution for unsupervised motor anomaly detection. EAAI, 162, 112446. (www)

  @article{ , abstract = {—Permanent magnet synchronous motors (PMSMs) are widely used in industrial applications but remain vulnerable to stator faults, such as inter-coil and inter-turn short circuits. Although recent deep learning-based fault detection methods have shown promise, they typically rely on large volumes of labelled fault data for training. To address this limitation, this paper proposes a novel unsupervised fault detection framework, termed Deep Adaptive Wavelet Autoencoder (DAWA) with Mutually Independent Empirical Cumulative Distribution (MIECD), specifically designed for PMSM fault detection. DAWA utilizes convolutional neural networks to learn adaptive wavelet filters through fast discrete wavelet transform, allowing for fully learnable, threshold-free extraction of fine-grained signal patterns. The resulting latent features are then mapped by MIECD into a mutually independent space via independent component analysis (ICA). Without assuming any prior data distribution, MIECD estimates empirical cumulative distributions (ECDs), computes tail probabilities across dimensions, and aggregates them into a unified anomaly score. Experimental results on motor vibration datasets demonstrate the effectiveness of the proposed method, showing average accuracy improvements of 15.85 % for Interturn and 15.16 % for Intercoil fault detection compared to conventional data-driven baselines across various operating conditions.}, archive = {J_EAAI}, author = {Pinze Ren and Ning Zhu and Dandan Peng and Liyuan Ren and Huan Wang}, doi = {10.1016/j.engappai.2025.112446}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112446}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep adaptive wavelet autoencoder with mutually independent empirical cumulative distribution for unsupervised motor anomaly detection}, volume = {162}, year = {2025}, }
• (2025). A reliable degradation prediction method for proton exchange membrane fuel cells based on uncertainty bayesian self-attention. EAAI, 162, 112444. (www)

  @article{ , abstract = {Health state prediction of Proton Exchange Membrane Fuel Cells (PEMFCs) is a critical technology to ensure their long-term reliable operation. Prediction accuracy directly influences the effectiveness of maintenance strategies and risk management. However, existing PEMFC degradation prediction methods based on Recurrent Neural Networks (RNNs) or Transformer architectures mostly focus on point estimation while neglecting uncertainty quantification. This limitation makes it difficult to assess the confidence level of predictions in practical engineering applications, reducing the models' reliability in decision support. To address this issue, this paper proposes a novel Bayesian Patch Time Series Transformer (B-PatchTST) method. By deeply integrating Bayesian variational inference with time series patch modeling, the method enables probabilistic prediction of PEMFC degradation trajectories and disentangled analysis of uncertainty sources. Unlike traditional Bayesian Neural Networks (BNNs) that primarily apply Bayesian modeling to fully connected layers, B-PatchTST introduces a Bayesian Self-Attention Mechanism, which models epistemic uncertainty in three stages: patch embedding, uncertainty-aware self-attention computation, and adaptive regularization. This design significantly enhances the credibility of the model. Extensive experiments on the fuel cell datasets demonstrate the proposed method's outstanding performance. It achieves an average reduction of 36.31 % in root mean square error and an average compression of 83.39 % in the 95 % confidence interval, significantly outperforming existing methods. This approach offers a trustworthy basis for predictive maintenance in PEMFC systems, promoting a shift from “experience-based maintenance” to “reliable prognostics” in hydrogen energy applications.}, archive = {J_EAAI}, author = {Mengyu Liu and Zhe Cheng and Yu Yang and Niaoqing Hu and Guoji Shen and Yi Yang}, doi = {10.1016/j.engappai.2025.112444}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112444}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A reliable degradation prediction method for proton exchange membrane fuel cells based on uncertainty bayesian self-attention}, volume = {162}, year = {2025}, }
• (2025). WeedNet-X: A lightweight field weed detection algorithm. EAAI, 162, 112441. (www)

  @article{ , abstract = {Accurately distinguishing field weeds from crops and locating weed positions are critical prerequisites in automated weed control operations. However, weed detection and localization in unstructured field environments with complex lighting remains challenging. Firstly, data-driven deep learning based algorithms usually have a high dependence on a large number of training samples, and there are huge differences between field weeds and crops in different regions, growth cycles, and types. In addition, the conflict between hardware performance and computation cost makes it difficult for existing weed detection algorithms to maintain both detection accuracy and speed on low-performance platforms. All these problems increase the difficulty of detection. To solve the above problems, we first construct a medium-to-large weed dataset using an open-source agricultural image dataset and collect field data. Subsequently, we have proposed a lightweight weed detection algorithm using the ShuffleNetv2 network as the backbone network, with a multi-scale pyramid network, and the overall network algorithm is named WeedNet-X. The number of model parameters and the computational volume of the algorithm are only 0.57 million and 0.48 Giga floating point operations (GFLOPs), respectively. On the two constructed datasets, the mean Average Precision (mAP) of the algorithm can reach 86.31 % and 80.98 %, respectively, which are improved by 0.61 % and 3.10 % compared to the baseline model. Finally, the hardware and software systems for weed detection verify the excellence of the proposed algorithm in terms of practical performance.}, archive = {J_EAAI}, author = {Yong Li and Ao Ke and Zhiqiang Guo and Qingji Tan and Jingchao Yang}, doi = {10.1016/j.engappai.2025.112441}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112441}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WeedNet-X: A lightweight field weed detection algorithm}, volume = {162}, year = {2025}, }
• (2025). Data-driven robust intertwined wheat supply chain: Redesigning a viable network for long-term geopolitical conflicts. EAAI, 162, 112440. (www)

  @article{ , abstract = {The recent human-made conflict in 2022 severely damaged Ukraine's infrastructure, causing significant instability in the food supply chain. This crisis was further exacerbated by trade bans imposed on another major global wheat exporter. Since wheat production and export are intrinsically linked, particularly in times of crisis, it is essential to adopt the concept of an intertwined supply chain. Accordingly, this study proposes an intertwined supply chain framework for the production and export of wheat during long-term disruptions. To enhance the viability of this intertwined system, the study introduces three key strategies. First, it addresses long-term disruptions and operational risks by employing redundancies and data-driven robust optimization techniques, where uncertainty sets are generated using a support vector clustering model. Second, the proposed supply chain accounts for freshwater resource limitations by integrating water resilience measures. Third, as the framework operates within a global context, it incorporates a comprehensive model that considers exchange rates, taxation, foreign demand points, and international trade responsibilities. To optimize these strategies, two multi-objective optimization models are developed and solved using an epsilon-constraint method. A cardinality-based measure is introduced to efficiently represent the Pareto front, offering decision-makers valuable insights into non-dominated solutions. The results are divided into analyses of wheat production, export, and their combined network. Individual analyses assess network setup, viability, and uncertainty control, while the integrated analysis examines sensitivity and interdependence. Overall, improving water use, managing risks, and designing a resilient, interconnected system can greatly strengthen the wheat supply chain during long-term crises.}, archive = {J_EAAI}, author = {Hani Gilani and Mehrdad Mohammadi and Tom Van Woensel and Hadi Sahebi}, doi = {10.1016/j.engappai.2025.112440}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112440}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven robust intertwined wheat supply chain: Redesigning a viable network for long-term geopolitical conflicts}, volume = {162}, year = {2025}, }
• (2025). Carbon trading price prediction with spikes: A novel hybrid model framework using heuristic multi-head attention convolutional bidirectional recurrent neural network. EAAI, 162, 112438. (www)

  @article{ , abstract = {Accurate forecasting of carbon trading prices (CTPs) with spikes is crucial for developing carbon emission reduction policies and planning corporate investments. However, most existing CTP approaches usually focus on designing a cutting-edge model without considering spike prediction. Therefore, this paper presents a novel heuristic optimization-based hybrid model framework for CTP prediction with spikes. First, random forest is exploited to identify the relevant features of spikes and non-spikes for CTPs, and categorical boosting is employed to predict the spike occurrences of CTPs. Then, a novel hybrid model based on multiple linear regression, categorical boosting, and two dimensions convolutional neural network and bidirectional gated recurrent unit with multi-head regularized attention mechanism (2DCNN-BiGRU-MRA) is proposed to predict spikes and non-spikes for CTPs. In this model, multiple linear regression and categorical boosting are respectively applied to capture the linear and complex nonlinear features of the CTPs, in which their prediction results and deviations are integrated into the 2DCNN-BiGRU-MRA model as relevant features. The proposed 2DCNN-BiGRU-MRA can learn the spatiotemporal features and enhance representation capabilities by introducing 2DCNN, BiGRU, and MRA, thereby improving the accuracy of CTP prediction. In addition, to construct appropriate model hyperparameters of 2DCNN-BiGRU-MRA, the strength honey badger algorithm based on the adaptive momentum estimation is proposed to optimize the hyperparameters of 2DCNN-BiGRU-MRA. Finally, the proposed framework is tested on the actual data of European Union emissions trading and the carbon market in Hubei, China, and case studies have confirmed the superiority and achievable local interpretability of the proposed hybrid model framework.}, archive = {J_EAAI}, author = {Rongquan Zhang and Siqi Bu and Gangqiang Li and Min Zhou}, doi = {10.1016/j.engappai.2025.112438}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112438}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Carbon trading price prediction with spikes: A novel hybrid model framework using heuristic multi-head attention convolutional bidirectional recurrent neural network}, volume = {162}, year = {2025}, }
• (2025). A real-time spatiotemporal error compensation framework for face gear grinding. EAAI, 162, 112429. (www)

  @article{ , abstract = {Geometric and thermal errors critically affect the precision of face gear grinding, yet current modeling approaches are computationally intensive and lack real-time adaptability. This study proposes a real-time spatiotemporal error compensation framework for face gear grinding. A closed-loop feedback mechanism is introduced to adaptively update compensation intensity based on residual error feedback, ensuring robustness and efficiency under fluctuating machining conditions. Moreover, a novel spatial-temporal thermal error model is developed by integrating Taylor-graph convolutional network and modified-long short term memory network to capture both node-level spatial fusion and long-term temporal dependencies. High-order terms in geometric error modeling are eliminated using a vector decomposition and truncation-based approach, significantly reducing computational complexity. Furthermore, a high-efficiency multi-source error-tooth flank mapping model is developed based on vector decomposition and truncation function methods, enabling accurate prediction with reduced computational cost. To identify dominant error contributors, an improved Morris-based sensitivity analysis method is integrated, distinguishing geometric and thermal errors affecting tooth flank deviation. Experimental results demonstrate sub-65 ms real-time response, 24.2 μm maximum error reduction, and robust adaptability under fluctuating machining conditions. Compared with recent gear-flank compensation studies, the proposed closed-loop framework achieves a 63.4 % reduction in maximum normal flank error under real machining and <65 ms response latency. This level is comparable to reported reductions based on grid-aggregated metrics in spiral bevel gears (76.82 % reduction of the sum of absolute grid errors), while additionally ensuring real-time, delay-aware execution. These findings validate the proposed system's potential for precision, real-time compensation in multi-axis manufacturing environments.}, archive = {J_EAAI}, author = {Jialan Liu and Chi Ma and Mingming Li and Jialong He and Giovanni Totis and Chunlei Hua and Gangwei Cui and Liang Wang and Ruijun Xue and Zhi Tan and Jun Yang and Kuo Liu and Yuansheng Zhou and Jianqiang Zhou and Xiaolei Deng and Shengbin Weng}, doi = {10.1016/j.engappai.2025.112429}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112429}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A real-time spatiotemporal error compensation framework for face gear grinding}, volume = {162}, year = {2025}, }
• (2025). Damage and energy dissipation prediction and multi-objective optimization design of latticed concrete-filled steel tube column-composite box girder joints based on extreme gradient boosting. EAAI, 162, 112426. (www)

  @article{ , abstract = {To optimize joint performance, a finite element (FE) model is developed based on low-cycle reciprocating load tests of latticed concrete-filled steel tubular (CFST) column-composite box girder joints. The FE-predicted hysteresis curves are compared with test results to verify model accuracy, and a data set is established accordingly. Extreme Gradient Boosting (XGBoost) algorithm is used for training and prediction, and compared with the traditional machine learning (ML) algorithm, the superiority of the XGBoost algorithm is manifested. The XGBoost algorithm is then used to predict the damage and energy values of the joint under more different parameter combinations, with the largest ratio of damage value to energy dissipation value selected as the optimal combination of the joints within the variation range of the six parameters. The results show that the FE model correlates well with the test results and can therefore be used to generate a data set. The prediction accuracy of XGBoost algorithm has high accuracy of more than 99 % in predicting damage and energy dissipation values and can thus be used for joint prediction research. Compared with other ML algorithms, XGBoost has the best prediction performance and superiority. Within the variation range of the six parameters, the ratio of damage value to energy dissipation value is the largest when the concrete strength, longitudinal bar diameter, concrete slab thickness, box girder strength, axial compression ratio, and transverse stiffener strength are respectively 30 Mega Pascal (MPa), 12 mm (mm), 90 mm, 390 MPa, 0.3, and 455 MPa.}, archive = {J_EAAI}, author = {Zhi Huang and Xin Deng and Juan Chen and Xiang Li and Lizhong Jiang and Yohchia Frank Chen and Yuner Huang and Lin Chen}, doi = {10.1016/j.engappai.2025.112426}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112426}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Damage and energy dissipation prediction and multi-objective optimization design of latticed concrete-filled steel tube column-composite box girder joints based on extreme gradient boosting}, volume = {162}, year = {2025}, }
• (2025). FDFNet: Feature-decision dual fusion network for intelligent fault diagnosis of rotating machinery under varying speed conditions. EAAI, 162, 112423. (www)

  @article{ , abstract = {Fault diagnosis is a critical technique for enhancing the reliability and security of rotating machinery. Existing diagnosis methods still have restricted generalization performance under speed variations owing to inadequate utilization of multisensor information. To address this problem, a novel feature-decision dual fusion network is proposed for fault diagnosis of rotating machinery under varying speed conditions. First, for each sensor, the frequency information learner is built to simultaneously extract global and local frequency domain features using global and local feature encoders. These features are then fused through a cross-attention mechanism to generate a sensor-specific initial classification decision. Subsequently, these individual sensor-wise decisions are fed into the decision dynamic ensemble to yield final fault diagnosis result. Moreover, an adaptive optimization strategy is designed to guide model learning generalizable features by flexibly adjusting the sensor loss weights during training. Finally, the effectiveness of the proposed method is validated on bearing and gearbox datasets. Experimental results demonstrate that the proposed method exhibits superior generalization and robustness for fault diagnosis under varying speed conditions.}, archive = {J_EAAI}, author = {Qi Deng and Zuoxiu Zhang and Xuyuan Tu and Zimuzhi Wang and Jun Wu}, doi = {10.1016/j.engappai.2025.112423}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112423}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {FDFNet: Feature-decision dual fusion network for intelligent fault diagnosis of rotating machinery under varying speed conditions}, volume = {162}, year = {2025}, }
• (2025). Lightweight deep learning based solar cells defect detection using electroluminescence images. EAAI, 162, 112421. (www)

  @article{ , abstract = {Solar cells are the fundamental core energy harvesting components in photovoltaic (PV) power generation stations. In view of the capability of detecting the invisible defects, the electroluminescence (EL) imaging is broadly used in the production lines of solar cells, based on which the deep learning technique is introduced to implement automatic defect detection and classification. However, the current deep learning models feature high complexity and require much computation resources, which are difficult to deploy in edge devices for real time applications. To tackle this issue, we proposed a novel lightweight and high-precision deep learning model named Cross Stage Partial Photovoltaic-You Only Look Once (CSPV-YOLO) based on the deep learning framework You Only Look Once v5 (YOLOv5) to enable the real-time solar cell defect detection. Firstly, a new module Cross Stage Partial C5 (CSPC5) is proposed to replace the initial C3 module in the YOLOv5 network to enhance the network accuracy in recognizing different types of defects. Secondly, a novel Spatial Pyramid Pooling with Cross Stage Partial (SPPFCSP) module is designed to replace the original Spatial Pyramid Pooling Fast (SPPF), which boosts the network feature extraction capabilities from defect targets at multiple scales and facilitates a more efficient integration of multiscale features. Finally, the original loss function of YOLOv5 is replaced by the Scylla intersection over union (SIoU) function to optimize the training model. The proposed models have been validated and intensively compared with many other state-of-the-art models on two public datasets. Firstly, results of experiments on the public Pascal Visual Object Classes (PASCAL VOC) 2007 datasets demonstrate that the proposed SPPFCSP block is obviously superior to other Spatial Pyramid Pooling blocks for the most state-of-the-art YOLO detectors, which can significantly improve the detection accuracy. The comparison results of experiments on the public Photovoltaic Electroluminescence Anomaly Detection Dataset (PVEL-AD) that includes 12-class defects obviously indicate that the proposed CSPV-YOLO model is better than many state-of-the-art models and achieves 91.5 % average precision (AP) and frames per second (FPS) of 177.8 on with only 2.2 million (M) parameters. Hence, it is suitable for the deployment on edge devices for real-time applications.}, archive = {J_EAAI}, author = {Zhicong Chen and Tianxiang Chen and Haoxin Zheng and Lijun Wu and Shuying Cheng and Peijie Lin}, doi = {10.1016/j.engappai.2025.112421}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112421}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight deep learning based solar cells defect detection using electroluminescence images}, volume = {162}, year = {2025}, }
• (2025). LGP-DETR: An end-to-end object detection model for surface defect detection in light guide plates. EAAI, 162, 112416. (www)

  @article{ , abstract = {In this paper, we propose Light Guide Plate-DETR (LGP-DETR), an end-to-end object detection model tailored for identifying surface defects in light guide plates (LGPs). To address challenges such as low contrast, small target size, and complex backgrounds in industrial settings, LGP-DETR integrates three key components: Deformable Transformer Fusion Layer (DtransFusion), a deformable attention-based fusion module for capturing multi-scale features; Upsampling by Dynamic Sampling (DySample), a dynamic upsampling strategy for edge detail preservation; and OrthoC3, a channel attention module that suppresses background noise through orthogonal feature enhancement. We adopt FasterNet as a lightweight convolutional backbone to achieve a balance between accuracy and efficiency. Experimental results on a real-world LGP defect dataset demonstrate that LGP-DETR achieves a mean Average Precision (mAP) of 97.9 % and inference speed of 60 frames per second (FPS), significantly outperforming existing models. Furthermore, generalization tests on a fiberglass fabric defect dataset confirm the model's adaptability to different industrial domains. These findings validate the practical applicability and robustness of the proposed deep learning framework for industrial visual inspection.}, archive = {J_EAAI}, author = {Shuangning Liu and Cunling Liu and Junfeng Li}, doi = {10.1016/j.engappai.2025.112416}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112416}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {LGP-DETR: An end-to-end object detection model for surface defect detection in light guide plates}, volume = {162}, year = {2025}, }
• (2025). Efficient interactive segmentation of three-dimensional gaussians with optimal view selection. EAAI, 162, 112413. (www)

  @article{ , abstract = {Three-dimensional (3D) scene representation has advanced rapidly in recent years, drawing the focus of more researchers. One of the main challenges for researchers is quickly and accurately segmenting 3D objects. Previous work has achieved excellent segmentation accuracy, but retraining requires a significant amount of time. Additionally, most methods fail to provide users with an efficient and convenient segmentation experience. To address these issues, we present Efficient Interactive Segmentation of 3D Gaussians (EISG), an efficient interactive segmentation method that eliminates the need for lengthy retraining. We first design an optimal view selection (OVS) method. This method uses 3D Gaussian entropy and image uncertainty to evaluate the quantity of view information. OVS helps users quickly select the optimal segmentation view, thereby enhancing interaction efficiency. Secondly, we use projection to find the target foreground rapidly and then segment the approximate objects using a clustering algorithm. Thirdly, we design a spatial-color background filter (SCBF) using the depth and color of 3D Gaussians. SCBF enables precise segmentation without needing retraining. Our method has been systematically tested on multiple datasets. Compared to other methods, the results demonstrate that EISG achieves ideal accuracy while significantly reducing processing time.}, archive = {J_EAAI}, author = {Yongtang Bao and Chengjie Tang and Yuze Wang and Yutong Qi and Ruijun Liu}, doi = {10.1016/j.engappai.2025.112413}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112413}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient interactive segmentation of three-dimensional gaussians with optimal view selection}, volume = {162}, year = {2025}, }
• (2025). Adversarial black-box attack and defense for convolutional neural network-based power quality disturbance classification. EAAI, 162, 112411. (www)

  @article{ , abstract = {Correctly identifying power quality disturbance (PQD) is crucial for the proper functioning of power systems. Deep learning (DL) techniques have been widely used for PQD classification due to their excellent performance. However, DL models are susceptible to adversarial attacks, posing a serious security threat to DL-based PQD classification systems. This issue has received limited attention in current research. In this study, we first utilize a convolutional neural network (CNN) to recognize various types of PQD signals. To evaluate model robustness, we introduce a black-box attack method for PQD classification based on the variance-tuning momentum iterative fast gradient sign method (VMI-FGSM). VMI-FGSM integrates a variance tuning method into the iterative process of the momentum iterative fast gradient sign method (MI-FGSM) , thereby producing more transferable adversarial PQD signals. To defend against such attacks, we propose a perturbation removal defense based on a generative adversarial network (PRD-GAN). This approach is capable of removing perturbations from adversarial PQD signals before they are recognized by the target classification model. Experiments demonstrate that VMI-FGSM produces adversarial perturbations that are nearly identical to those of the advanced MI-FGSM, but its adversarial examples are significantly more effective at misleading the target CNN model. Furthermore, the proposed PRD-GAN effectively reconstructs adversarial PQD signals into clean forms under various black-box attack intensities and outperforms the multi-level denoising autoencoder (ML-DAE) in defense performance due to its superior reconstruction capability.}, archive = {J_EAAI}, author = {Xiudong Zhang and Congmei Jiang and Mingbiao Yu and Xiankui Wen and Jing Zhang and Na Rong and Song Han}, doi = {10.1016/j.engappai.2025.112411}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112411}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adversarial black-box attack and defense for convolutional neural network-based power quality disturbance classification}, volume = {162}, year = {2025}, }
• (2025). Adaptive resource management in dynamic Cyber–Physical systems using artificial intelligence. EAAI, 162, 112409. (www)

  @article{ , abstract = {Effective resource management is crucial for the operation of dynamic Cyber–Physical Systems (CPS), yet traditional static or rule-based approaches often fail to handle their inherent complexity. This paper presents a novel Artificial Intelligence (AI)-driven framework for adaptive resource management, defined as the capability to autonomously adjust resource allocation by proactively forecasting future demands and dynamically optimizing decisions in real-time. The framework integrates a suite of AI techniques, including time-series models like Long Short-Term Memory (LSTM), chosen for their ability to capture complex temporal dependencies, for demand prediction. For resource allocation, it employs advanced actor–critic reinforcement learning (RL) algorithms like Proximal Policy Optimization (PPO) and Deep Deterministic Policy Gradient (DDPG), selected for their stability and efficiency in complex decision-making tasks. Performance was rigorously evaluated in a simulated dynamic environment. Experimental results demonstrate that combinations leveraging LSTM’s predictive accuracy with the robust optimization of PPO and DDPG achieve superior performance and stability. Specifically, the LSTM+DDPG and LSTM+PPO configurations yielded the highest average rewards (0.964 and 0.942, respectively), significantly outperforming the fixed-strategy baseline (0.497) and other AI pairings. Furthermore, the feasibility of training prediction models in a distributed manner via Federated Learning (FL) is successfully demonstrated. This research highlights that a synergistic integration of suitable AI predictors and advanced RL agents provides a powerful and resilient solution for resource management in dynamic CPS.}, archive = {J_EAAI}, author = {Xiaofei Zhao and Fangling Guo and Amin Huang and Jieqiong Ding and Chi Yan and Wei Yuan and Yunqi Su and Quanzhou Li and Qianggang Zhang}, doi = {10.1016/j.engappai.2025.112409}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112409}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive resource management in dynamic Cyber–Physical systems using artificial intelligence}, volume = {162}, year = {2025}, }
• (2025). A novel prior knowledge-based and texture-enhanced network for few-shot industrial defect segmentation. EAAI, 162, 112408. (www)

  @article{ , abstract = {Deep learning-based segmentation techniques have demonstrated significant potential in defect detection, which is vital for product quality. However, the existing models tend to specialize in detecting specific defect types, reducing their adaptability to a wider range of product defects, most existing methods rely on multi-scale or prototype learning to extract defect features, but still struggle with complex backgrounds, various interferences, and large intra-class variations. Additionally, the rarity of certain defects limits the availability of training samples. Herein, an innovative segmentation network, the Prior Knowledge-based and Texture-Enhanced Network (PTNet), is designed for few-shot industrial segmentation. The model is mainly composed of a self-guidance branch, a cross-guidance branch, and a texture enhancement module, enabling generalization across defect types with minimal labeled samples. Self-guidance extracts prior knowledge from the query image, while the cross-guidance branch extracts prior and prototype features from the masked support image. The texture information in the low-level features of the backbone is then enhanced by proposed texture enhancement module (TEM). Finally, the enhanced low-level texture information are fused with high-level semantic features, allowing the network to fully exploit both local details and global context before being decoded to restore the original image size. This enables the model to handle complex textures and generalize to unseen defect types. Extensive experimental results validate the effectiveness of the proposed modules. State-of-the-art performance is achieved in few-shot defect segmentation, with notable improvements in mean Intersection over Union (mIoU) of 46.05 % and 46.98 % under 1-shot and 5-shot conditions, respectively.}, archive = {J_EAAI}, author = {Xingyue Liu and Qian Wu and Yahui Cheng and Guojun Wen}, doi = {10.1016/j.engappai.2025.112408}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112408}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel prior knowledge-based and texture-enhanced network for few-shot industrial defect segmentation}, volume = {162}, year = {2025}, }
• (2025). A novel reinforcement learning-based approach for optimal control: An application to multi-tank water systems. EAAI, 162, 112407. (www)

  @article{ , abstract = {The optimization of control actions is a critical challenge in industrial systems, especially when dealing with complex and unknown dynamics. Data collected from the environment enables the application of reinforcement learning techniques, which let the controller learn a policy based on data. This work proposes a novel model-free reinforcement learning approach that consists of a value iteration algorithm based on separate policy evaluation and policy improvement phases to provide an accurate control policy estimation. The proposed approach addresses tracking control for quadruple-tank water systems while obtaining minor tracking errors and faster transient responses. The results from the case study reveal better accurate estimation of the value function, up to 86.13% mean improvement in tracking accuracy and faster responses compared to existing methods. Therefore, the proposed approach demonstrates advantages in optimizing control performance and stands as a promising control method for industrial applications.}, archive = {J_EAAI}, author = {Eva Masero and Giacomo Mussita and Alessio La Bella and Riccardo Scattolini}, doi = {10.1016/j.engappai.2025.112407}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112407}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel reinforcement learning-based approach for optimal control: An application to multi-tank water systems}, volume = {162}, year = {2025}, }
• (2025). Automatic weighted ensemble learning for ballast resistance estimation driven by vehicle-ground information fusion. EAAI, 162, 112406. (www)

  @article{ , abstract = {Health management of transmission parameters for railway signal equipment is a key link between intelligent operation and maintenance. As a core parameter of track circuits, ballast resistance significantly affects signal transmission. To accurately and reliably assess its health state, an ensemble learning algorithm (ELA) is introduced, tackling deviations of appraisal decision boundaries. Focusing on issues of complex weight calculation, model homogenization, and severe overfitting in ELA, an integration model based on an automatic weight allocation strategy (AWAS) is innovatively proposed, constructing a method for resistance estimations driven by information fusion, while maximizing its generalization ability. Firstly, for deterioration mechanism analysis of ballast resistance, a transmission state model for vehicle-ground collaboration is established, completing extractions of evolutionary rules. Secondly, the improved ELA leverages heterogeneous classifier optimization and automatic weighted soft voting, with its core ensemble strategy employing a secondary learner to map the fused datasets. Then, by means of data mining techniques, interpolation and denoising algorithms are applied to implement data preprocessing, facilitating the effective fusion of heterogeneous vehicle-ground information. Finally, based on occurrence of adverse conditions, an appropriate particle size is set to achieve state warning. The results indicate that the proposed AWAS for ballast resistance calculations can achieve 98.52 % testing accuracy and outperforms others.}, archive = {J_EAAI}, author = {Conghui Wang and Shiwu Yang and Chang Liu}, doi = {10.1016/j.engappai.2025.112406}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112406}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic weighted ensemble learning for ballast resistance estimation driven by vehicle-ground information fusion}, volume = {162}, year = {2025}, }
• (2025). Future of humanity in an artificial intelligence centric world. EAAI, 162, 112405. (www)

  @article{ , abstract = {This scholarly article rigorously investigates the transformative and disruptive roles of artificial intelligence (AI) in influencing the trajectory of human society. By concentrating on three fundamental sectors—healthcare, finance, and education—it evaluates the ways in which AI augments operational efficiency, facilitates intricate decision-making processes, and introduces innovative capabilities such as personalized medicine and automated financial systems. Concurrently, the analysis underscores urgent ethical dilemmas, encompassing algorithmic bias, accountability deficiencies, data privacy vulnerabilities, and workforce displacement. Employing real-world examples such as Deepfakes, and Neuralink, the article contextualizes emerging challenges within a dynamic socio-technical framework. The research offers a cohesive conceptual model that amalgamates technical, ethical, and governance aspects of AI, while presenting policy recommendations designed to promote transparency, equity, and human-centered AI development. The study emphasizes the necessity for reliable AI systems that humans can trust. The conclusions accentuate the immediate necessity for robust regulatory frameworks and sector-specific ethical supervision to ensure that advancements in AI are harmonized with the well-being of society.}, archive = {J_EAAI}, author = {Sunil Baloda and Monika Sharma and Mukesh Kumar}, doi = {10.1016/j.engappai.2025.112405}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112405}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Future of humanity in an artificial intelligence centric world}, volume = {162}, year = {2025}, }
• (2025). A novel federated deep learning for intrusion detection in smart grid cyber-physical systems. EAAI, 162, 112404. (www)

  @article{ , abstract = {The fusion of sophisticated computational, communicative, and physical elements in Smart Grid Cyber-Physical Systems (SGCPS) has greatly improved the efficiency and reliability of power grids. However, this complexity introduces enhanced cybersecurity risks, evidenced by significant cyberattacks on the Ukrainian power grid during 2015 and 2016. Despite progress in Artificial Intelligence (AI)-driven security solutions for SGCPS, practical deployment of these technologies is often limited due to a lack of high-quality attack data and owners’ hesitance to distribute sensitive details. This paper introduces an innovative strategy to fortify SGCPS against diverse network threats via a comprehensive intrusion detection system. We present a deep learning model leveraging a temporal convolutional network with multi-feature integration, aimed at robust threat identification. We also propose a federated learning framework enabling various SGCPS to jointly develop an extensive intrusion detection model, ensuring data privacy. Moreover, we incorporate a gradient compression technique utilizing the Long Short Term Memory- β -Total Correlation Variational Autoencoder (LSTM- β -TCVAE) model to enhance and secure model parameters throughout the training phase. Thorough experimental validations confirm the efficacy of our method in recognizing multiple cyber threat types to SGCPS and its advantages over current methods.}, archive = {J_EAAI}, author = {Rong Xie and Bin Wang and Xin Xu}, doi = {10.1016/j.engappai.2025.112404}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112404}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel federated deep learning for intrusion detection in smart grid cyber-physical systems}, volume = {162}, year = {2025}, }
• (2025). Tunnel-YOLO: An improved you only look once algorithm for real-time shield tunnel lining leakage detection. EAAI, 162, 112403. (www)

  @article{ , abstract = {Detecting crack leakages in shield tunnels is crucial for ensuring structural safety and extending service life, as traditional detection methods are limited by high subjectivity and low accuracy. To address these limitations, this paper proposes Tunnel-YOLO, an improved object detection algorithm based on You Only Look Once version 8 (YOLOv8). This algorithm replaces standard convolutional blocks with a novel Receptive Field Channel Attention Convolution (RFCAConv) module, which leverages dynamic receptive fields to enhance feature capture at different scales. We also introduce a C2f_SGE module, integrating the Spatial Group-wise Enhance (SGE) attention mechanism into the C2f (CSPNet with 2 convolutions) block to significantly improve feature extraction while suppressing background interference. Furthermore, an Edge Feature Enhancement Detection Head (EFE-Head) incorporates deconvolution layers to enhance fine-grained details for more precise boundary localization. To better accommodate the shape-sensitive detection task, our LeShape-IoU (Intersection over Union) loss function is designed to focus on the shape and scale characteristics of target bounding boxes. Experimental results on a public, real-world dataset demonstrate that Tunnel-YOLO significantly outperforms the baseline, increasing Recall, Precision, and mean Average Precision at 0.5 IoU (mAP50) by 15.7%, 10.3%, and 14.8%, respectively. Comparative analysis with other mainstream algorithms further validates the effectiveness and superiority of the proposed Tunnel-YOLO.}, archive = {J_EAAI}, author = {Ruijun Yang and Hao Zhang}, doi = {10.1016/j.engappai.2025.112403}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112403}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tunnel-YOLO: An improved you only look once algorithm for real-time shield tunnel lining leakage detection}, volume = {162}, year = {2025}, }
• (2025). Contrastive active adaptive partial label learning under class distribution mismatch. EAAI, 162, 112401. (www)

  @article{ , abstract = {Partial label learning is an important learning framework where each training sample is associated with a candidate label set and its ground-truth label is included in the candidate label set. Active partial label learning is a variation where training data consists of both labeled and unlabeled samples. However, there exists the problem of class distribution mismatch, wherein the unlabeled sample set contains many instances out of the target categories. In this paper, a contrastive active adaptive partial label learning method under class distribution mismatch which combines active partial label learning with contrastive coding is proposed. A novel active sample selection strategy is first established to use label propagation ability to measure the optimization ability of unlabeled samples to partially labeled samples. Furthermore, to solve the problem of class distribution mismatch, a joint query score based on contrastive coding is utilized to reduce the queries of unlabeled samples out of target categories. Finally, the above two indicators are combined adaptively to select the most valuable unlabeled samples in target categories for manual labeling and the selected samples will be added to the training sample set to train the new classifier. The effectiveness and efficiency of the method are evaluated by performing experiments on the datasets CIFAR10 and CIFAR100.}, archive = {J_EAAI}, author = {Aohan Zhang and Kezhen Dong and Hongying Zhang}, doi = {10.1016/j.engappai.2025.112401}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112401}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrastive active adaptive partial label learning under class distribution mismatch}, volume = {162}, year = {2025}, }
• (2025). A novel weight-optimized machine-learning hybrid model for daily river runoff prediction. EAAI, 162, 112396. (www)

  @article{ , abstract = {The daily runoff process has been characterized as nonlinear and unsteady due to the impacts of watershed precipitation and evaporation, vegetation coverage rate, reservoir operations and other human activities. In recent years, machine-learning (ML) models have been widely applied in the daily runoff predictions, but the robustness and effectiveness of individual ML model is always limited. A novel weight optimization scheme has been introduced to ML models to obtain accurate predictions of daily river runoff. Variational modal decomposition method is adopted in the dataset preprocessing, and the runoff prediction performance of various classic ML models, including Genetic Algorithm-Back Propagation neural network (GA-BP), Long Short-Term Memory network (LSTM), Elman neural network (Elman) and Genetic Algorithm-Support Vector Machine (GA-SVM) are subsequently evaluated. A particle swarm optimization (PSO) based weight optimization strategy is proposed to combine different types of ML models, thus more accurate and robust results could be obtained. The ten-fold cross-validation method has been adopted and the performance of the optimized hybrid models are further evaluated for different schemes. A case study at Hankou hydrological station demonstrates that root mean square error (RMSE) and mean absolute percentage error (MAPE) is improved by 35.7 %, 75.8 % respectively for the optimized hybrid model. The present study shares useful insights to the comprehensive optimization of various ML models in the intelligent management of water resources.}, archive = {J_EAAI}, author = {Zhonglian Jiang and Jianglong Ying and Zhen Yu and Xiao Chu and Chengqiang Yu}, doi = {10.1016/j.engappai.2025.112396}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112396}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel weight-optimized machine-learning hybrid model for daily river runoff prediction}, volume = {162}, year = {2025}, }
• (2025). Intelligent assessment of habitat quality based on multiple machine learning fusion methods. EAAI, 162, 112395. (www)

  @article{ , abstract = {Evaluating habitat quality can help balance the relationship between economic development and biodiversity conservation, and it serves as a foundation for constructing an ecological security pattern. However, research on the intelligent construction of habitat quality is limited. This study develops a comprehensive framework to assess habitat quality based on optimized machine learning methods. The findings of the research are as follows: (1) From the perspective of human-machine interactive interpretation, ensemble learning is used to enhance the performance of basic classifiers, resulting in a classification map with high precision and recall. (2) The particle swarm optimization (PSO) algorithm can improve the goodness of fit of the Extreme Gradient Boosting (XGBoost) inversion model by 4–5 %. (3) The habitat quality inversion method based on XGBoost-PSO has high credibility and application value, with its texture structure being the result of both expert experience and image information interaction. (4) The model demonstrates certain application potential in downscaling; under the seven-band perspective, the blue and near-infrared bands are the most important, while in the four-band perspective, green and near-infrared bands take precedence.}, archive = {J_EAAI}, author = {Kui Yang and Dongge Cui and Chengrui Wang and Qi Tang and Linguang Miao}, doi = {10.1016/j.engappai.2025.112395}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112395}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent assessment of habitat quality based on multiple machine learning fusion methods}, volume = {162}, year = {2025}, }
• (2025). Hierarchical semantics guided multi-scale correlation network for alignment-free red-green-blue and thermal salient object detection. EAAI, 162, 112394. (www)

  @article{ , abstract = {RGBT (red-green-blue and thermal) salient object detection (SOD) aims to identify and highlight the most visually salient objects in an image by leveraging the complementary information from both RGB and thermal (TIR) modalities. It is particularly effective for 24/7 intelligent surveillance and autonomous perception in smart city security and traffic monitoring, especially under low light and adverse weather. However, existing methods primarily rely on manually aligned datasets, which are limited in handling the challenges posed by unaligned multi-modal data in real-world applications. Furthermore, these methods usually extract complementary information from both modalities using fixed-size windows (Liuet al., 2022, Wanget al., 2024b). However, such fixed-size windows are not effective in dealing with unaligned multi-modal images due to spatial inconsistencies. Additionally, existing methods often use single-layer high-level feature to represent semantic information, which fails to fully exploit the complementary benefits of multi-level features, thereby reducing the effectiveness of semantic guidance. To address these challenges, we propose a Hierarchical Semantics guided Multi-scale correlation Network (HSMNet) for alignment-free RGBT SOD. A Hierarchical Semantic Fusion Module (HSFM) dynamically assigns weights to features from multiple levels, enabling adaptive fusion of multi-level semantic information. A Multi-scale Asymmetric Correlation Module (MACM) employs windows of various sizes to capture asymmetric correlations between unaligned multi-modal data, enhancing cross-modal complementary information extraction even when data are not perfectly aligned. We conduct extensive experiments on unaligned, weakly aligned and aligned RGBT SOD datasets, with results demonstrating that our method outperforms state-of-the-art algorithms, achieving superior accuracy and robustness in both unaligned and weakly aligned RGBT SOD scenarios.}, archive = {J_EAAI}, author = {Chengmei Han and Lei Liu and Kunpeng Wang and Fei Xie and Bing Wei}, doi = {10.1016/j.engappai.2025.112394}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112394}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical semantics guided multi-scale correlation network for alignment-free red-green-blue and thermal salient object detection}, volume = {162}, year = {2025}, }
• (2025). Scientific machine learning for generic compact model parameter extraction of nanoscale transistors. EAAI, 162, 112392. (www)

  @article{ , abstract = {This work presents a framework to automate modelcard extraction of industry-standard compact models. This framework presents a Scientific Machine Learning (ScML) approach capable of inverse modeling. It integrates a random forest model with an artificial neural network to produce efficient and precise regression results. This new method is used for parameter extraction in the standard compact models of the semiconductor device industry. Modeling of a multi-gate Field Effective Transistor (FET) with an industry-standard compact model like Berkeley Short-channel Insulated-Gate Field-Effect Transistor Model – Common Multi-Gate (BSIM-CMG) is taken as an example to illustrate and describe the framework and highlight its key advantages. Proposed framework is useful in numerous aspects; it holds vital principles of physics, avoids depending on massive datasets, and has a sparse architecture while avoiding accuracy trade-offs. The framework is tested on production-level experimental devices to evaluate the real-world performance. This framework significantly reduces the time and cost of parameter extraction for the Process Design Kits (PDKs) development. Therefore, this is of immediate importance for fabrication and Electronic Design Automation (EDA) industries.}, archive = {J_EAAI}, author = {Kumar Sheelvardhan and Surila Guglani and Abhilash Dubey and Shashank Dubey and Sindhu Ramaswamy and Vaidy Subramanian and Kassandra Anderson and Glenn Workman and Sourajeet Roy and Avirup Dasgupta}, doi = {10.1016/j.engappai.2025.112392}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112392}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scientific machine learning for generic compact model parameter extraction of nanoscale transistors}, volume = {162}, year = {2025}, }
• (2025). Health state estimation of retired batteries based on physical constraints. EAAI, 162, 112390. (www)

  @article{ , abstract = {With the growing use of retired lithium-ion batteries, accurately monitoring their health status has become increasingly important. This study proposed a method to estimate the health of retired batteries by embedding their capacity degradation characteristics directly into the loss function of a Bidirectional Long Short-Term Memory (BiLSTM) network, combined with a Physically Informed Neural Network (PINN) model. The model is developed by incorporating the dynamics of the solid electrolyte interface (SEI) membrane, which evolves as the lithium-ion poles of the retired battery move. By combining these dynamics with the governing equations of motion, a partial differential equation (PDE) is derived. This approach integrates physical constraints, data-driven learning, and PDEs into a composite loss function. The proposed method is validated on two different datasets under varying operating temperatures. The results show that the PINN-BiLSTM model achieves a Root Mean Square Percentage Error (RMSPE) of 0.024, representing a 9.67 % improvement over the PINN-LSTM. This adaptive PINN method offers highly accurate health state predictions across temperature variations, thus supporting the sustainable use of retired batteries in secondary applications and helping to mitigate energy scarcity.}, archive = {J_EAAI}, author = {Fei Xia and Qianwen Dong and Lin Xia and Zhenyi An and Ziyang Xia and Chunyang Gong}, doi = {10.1016/j.engappai.2025.112390}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112390}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health state estimation of retired batteries based on physical constraints}, volume = {162}, year = {2025}, }
• (2025). Adaptive cybersecurity in bio-medical modules: A dynamic programming neural network-based protection for markov-chain fabricated data attacks. EAAI, 162, 112389. (www)

  @article{ , abstract = {With the advances in communication technologies, remote monitoring and control of drug delivery are becoming prevalent in bio-medical engineering. In a chemotherapy system, the measurement signals can be wirelessly transferred to the control center by communication networks. Nevertheless, the avenues of communication might be jeopardized by false data injection (FDI), which poses significant risks to the security and stability of biomedical systems. In this work, a defense mechanism is developed to tackle the effect of FDI threats in the networked chemotherapy system. In particular, the effect of FDI attacks on the chemotherapy system is modeled by the Markov chain process. The proposed defense mechanism is designed in two parts: i ) a data-driven sliding mode observer (DDSMO) is utilized to identify the occurrence of cyber attacks in the tumor signals measured by the bio-sensor, and ii ) a mitigation scheme based on a dynamic rejection compensator (DRC) to compensate for the impact of cyber threats. In the mitigation phase, a goal representation heuristic dynamic programming (GrHDP) is adopted to adaptively adjust the parameters of DRC and to dynamically handle the cyber threats. The designed mitigation mechanism not only regulates the cancerous cells against cyber threats but also minimizes the side effects of drug delivery by regulating the output of normal cell and immune cell. Compared to prevalent methodologies, the proposed approach yields significant performance, including a 60.23 % improvement over the without protection, 37.48 % over the DDSMO-based model predictive controller (MPC), 35.95 % over the reinforcement learning (RL) based Kalman filter, and 70.44 % over the proportional integral (PI) based Kalman filter.}, archive = {J_EAAI}, author = {Mostafa Taheri and Juliang Yin and Zahra Rasooli Berardehi}, doi = {10.1016/j.engappai.2025.112389}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112389}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive cybersecurity in bio-medical modules: A dynamic programming neural network-based protection for markov-chain fabricated data attacks}, volume = {162}, year = {2025}, }
• (2025). Weighted multi-source domain unsupervised adaptive network for rotating machinery fault diagnosis based on dual adversarial. EAAI, 162, 112386. (www)

  @article{ , abstract = {Multi-domain adaptive methods are becoming a growing focus of fault diagnosis, which can provide enhanced data support for models using the feature information from various source domains. As a most commonly used method, unsupervised multi-domain adaptive methods (UMA) can eliminate the requirement for the label of the target domain samples. However, the neglect of contributions from different source domains to the target domain and insufficient utilization of diagnostic information from multiple source domains are the widely limitation of UMA. Therefore, a dual-adversarial weighted multi-source domain unsupervised adaptive network (DAWMUN) is proposed to utilize diagnostic information from multi-source domains and consider the contribution of different source domains. Firstly, the shared feature extractor and dual adversarial training with the domain adversarial modules between multi-source domains and source-target domains are used to enhance domain confusion between multi-source and target domains (MSTD). Secondly, based on Multiple Kernel Maximum Mean Discrepancy (MK-MMD), a novel weighting mechanism and the corresponding training framework are constructed to effectively reduce negative transfer. Finally, a novel weighted classifier is proposed to merge the outputs of multiple classifiers and synthesize the impact of each source domain. The performance of the DAWMUN is validated using a rotating machinery dataset across various transfer tasks under different rotational speed and load conditions. The experimental results demonstrate that the diagnostic accuracy using the proposed DAWMUN is superior to existing SSDA and MSDA methods, with the average accuracies of 98.53 % and 98.23 % across six tasks in two separate experimental setups. The comparison to the existing methods results that the DAWMUN still demonstrates superior performance with improvements of 2.54 % and 2.86 %, respectively.}, archive = {J_EAAI}, author = {Wenqi Wang and Zongzhen Zhang and Jinrui Wang and Baokun Han and Huaiqian Bao and Zhikang Fan and Rongkang Ge}, doi = {10.1016/j.engappai.2025.112386}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112386}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted multi-source domain unsupervised adaptive network for rotating machinery fault diagnosis based on dual adversarial}, volume = {162}, year = {2025}, }
• (2025). Twin-stream network: Enhancing wave prediction by capturing spatiotemporal information. EAAI, 162, 112385. (www)

  @article{ , abstract = {Wave prediction is a critical challenge in ocean and coastal engineering, particularly for understanding and mitigating the effects of sea waves on structures such as ships, offshore platforms, and coastal defenses. A novel machine learning model, Twin-Stream Network (TSNet), is proposed to enhance wave prediction accuracy by leveraging temporal and spatial dependencies in historical data. The TSNet model along with other baseline models are evaluated, in both single-point and multi-point forecasting tasks, by various performance metrics across different datasets including one-dimensional-linear, one-dimensional-nonlinear, two-dimensional-linear and two-dimensional-nonlinear water waves. The comprehensive comparative analysis demonstrates that the TSNet model outperforms others, especially in the multi-point forecasting task. This study provides a valuable insight into the effectiveness of machine learning approaches and highlights the potential of the accuracy improvement for wave prediction.}, archive = {J_EAAI}, author = {Junhao Xu and Zhongying Feng and Zhan Wang and Kun Zheng and Ruipeng Li}, doi = {10.1016/j.engappai.2025.112385}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112385}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Twin-stream network: Enhancing wave prediction by capturing spatiotemporal information}, volume = {162}, year = {2025}, }
• (2025). Clean-sample guided cross-modal retrieval with adaptive weighted contrastive learning. EAAI, 162, 112383. (www)

  @article{ , abstract = {Cross-modal retrieval enables efficient integration of information by linking different data modalities, such as images and text. As data volumes increase rapidly, the need for effective cross-modal interaction grows. Cross-modal hashing is favored for its low storage requirements and fast retrieval speed, but many existing methods depend on accurately labeled data, which can be subjective and expensive to obtain. To address this limitation, we propose Clean-guided Adaptive Weighted Contrastive Hashing (CAWCH), a novel framework designed to improve robustness against noisy labels. CAWCH incorporates two main components: a Gaussian Mixture Model (GMM)-based noise purifier that identifies reliable and noisy samples by modeling sample loss, and a contrastive learning strategy that selectively chooses positive samples and adaptively assigns weights based on multi-label similarity, considering both intra- and inter-modal relationships. Extensive experiments demonstrate that CAWCH significantly outperforms existing methods under noisy label conditions, highlighting its effectiveness and potential for real-world cross-modal retrieval applications.}, archive = {J_EAAI}, author = {Shuni Jiang and Zhixin Li}, doi = {10.1016/j.engappai.2025.112383}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112383}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Clean-sample guided cross-modal retrieval with adaptive weighted contrastive learning}, volume = {162}, year = {2025}, }
• (2025). Denoising of the magnetic flux leakage signal using dynamic feature fusion and a multi-scale autoencoder network. EAAI, 162, 112382. (www)

  @article{ , abstract = {Magnetic flux leakage (MFL) signal denoising is essential for the nondestructive inspection of oil and gas pipelines, where complex noise interference can severely degrade defect quantification accuracy. Traditional approaches such as mean filtering and wavelet transform offer limited suppression of multi-type mixed noise and often distort critical features, including defect peaks and valleys. Even deep learning–based MFL denoising methods struggle in scenarios with substantial signal-noise overlap due to inadequate feature extraction and limited adaptability.This work presents an advanced denoising framework that combines dynamic feature fusion with a multi-scale autoencoder network. The framework jointly exploits time- and frequency-domain signal components, employing an adaptive weighting mechanism for dynamic feature fusion. Parallel convolutional branches extract multi-scale features, improving the capture of both global structures and fine-grained details, while a Squeeze-and-Excitation (SE) channel attention mechanism enhances defect-sensitive features and suppresses noise. Extensive experiments demonstrate that the proposed model outperforms mean filtering, wavelet denoising, and a baseline autoencoder, achieving notable gains in signal-to-noise ratio (SNR), mean squared error (MSE), and signal similarity. Beyond superior noise suppression, the method preserves critical defect characteristics, providing a robust and reliable foundation for precise defect quantification in pipeline MFL inspection.}, archive = {J_EAAI}, author = {Lushuai Xu and Shaohua Dong and Haotian Wei and Feng Li and Pengkun Zhang and Cong Zuo and Mingxing Guo and Penghui Liao}, doi = {10.1016/j.engappai.2025.112382}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112382}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Denoising of the magnetic flux leakage signal using dynamic feature fusion and a multi-scale autoencoder network}, volume = {162}, year = {2025}, }
• (2025). An optimisation approach guided by crack variation mechanism in the informer prediction model. EAAI, 162, 112381. (www)

  @article{ , abstract = {Structural health monitoring (SHM) faces a fundamental challenge in reconciling predictive performance with physical interpretability for infrastructure diagnostics. Conventional deep learning (DL) approaches neglect essential mechanisms governing crack width variation—including thermal gradients, hysteretic responses, and phase-shifted correlations—limiting their reliability in real-world applications. To bridge this gap, we propose a mechanism-guided optimization (MGO) framework that integrates domain knowledge into the Informer architecture through physics-informed enhancements: auto-correlation modeling for capturing temperature-crack hysteresis, static gated fusion for multi-feature integration, and adaptive elastic net regularization for feature selection. Validated on cable-stayed bridge monitoring data, our framework achieves significant mean absolute error reductions (MAE) (5 %–60 %) and root mean square error reductions (RMSE) (10 %–55 %) versus baseline Informer across all cracks and prediction horizons, with diebold-mariano (DM) tests confirming statistical superiority in most cases. Crucially, it demonstrates superior precision relative to six state-of-the-art benchmarks across all evaluation scenarios. The ordinary least squares (OLS)-enhanced variant further delivers volatility reduction, while sensor failure tests establish quantifiable robustness benchmarks through MAE progression from 0.013 mm to 0.391 mm. This work establishes an interpretable, physics-grounded paradigm that explicitly links environmental drivers to structural degradation.}, archive = {J_EAAI}, author = {Xujia Liu and Youliang Ding and Fei Xu and Yichao Xu and Kang Yang}, doi = {10.1016/j.engappai.2025.112381}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112381}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An optimisation approach guided by crack variation mechanism in the informer prediction model}, volume = {162}, year = {2025}, }
• (2025). A lightweight segmentation model based on segment anything model for tongue image segmentation. EAAI, 162, 112379. (www)

  @article{ , abstract = {Tongue image segmentation plays a crucial role in intelligent of diagnosis of Traditional Chinese Medicine. Accurate, efficient, and lightweight tongue segmentation significantly improves both the quality and practical applicability of intelligent disease diagnosis models. To address this challenge, we propose TongueSAM_Lite, a lightweight and fully automated tongue image segmentation model. Based on the Segment Anything Model. Our approach employs knowledge distillation and parameter-efficient fine-tuning to develop a novel lightweight image encoder, high-parameter modules in the Vision Transformer are partially replaced with lightweight image modules, which facilitate the transfer of its feature extraction capabilities while accelerating inference speed and reducing computational resource requirements. Additionally, to eliminate manual annotation of tongue region bounding boxes, we integrate a YOLOX-based automatic Box-prompt generator, enabling end-to-end fully automated prompting and segmentation of tongue images. To validate our approach, various experiments were conducted in three datasets. The results show that compared to the original large-scale model of the Segment Anything Model, TongueSAM_Lite reduces the size of the model by 42.7% and shortens the inference time to 45.43% while retaining the near-complete segmentation accuracy of few-shot learning. TongueSAM_Lite achieves Mean Intersection over Union scores of 96.48%, 98.36%, and 97.53% in the three datasets, respectively, outperforming state-of-the-art segmentation methods. Further validation confirms that the YOLOX-based prompt encoder yields optimal performance for the generation of tongue image bounding boxes. Our proposed approach provides new research insights to advance tongue diagnosis technology of Traditional Chinese Medicine. All codes in this article are available at https://github.com/ruanqunsheng/TongueSAM_Lite .}, archive = {J_EAAI}, author = {Qunsheng Ruan and Shan Cao and Zhirong Luo}, doi = {10.1016/j.engappai.2025.112379}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112379}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight segmentation model based on segment anything model for tongue image segmentation}, volume = {162}, year = {2025}, }
• (2025). Exploring cross-branch information for semi-supervised remote sensing object detection. EAAI, 162, 112378. (www)

  @article{ , abstract = {Semi-supervised object detection (SSOD) provides a promising solution to mitigate the annotation costs in remote sensing applications. Mainstream teacher-student based SSOD methods leverage unlabeled images through pseudo labeling, and their effectiveness is fundamentally limited by the inevitable noise in pseudo labels, particularly for remote sensing (RS) scenarios with complex backgrounds and dense, multi-scale and oriented objects. Current methods primarily focus on reducing pseudo label noise through category, scale and Intersection over Union information mining, as well as designing fine-grained confidence thresholding strategies. However, the inherent discrepancy between classification and localization reliability is neglected. In this study, with analyzing the characteristic discrepancies between the classification and localization branches, We propose artificial intelligence (AI) methodological innovation method named cross-branch information incorporation method (i.e., CBI-SSOD) to utilize these discrepancies to assist the training of the classification branch, and thus improve the performance of SSOD methods. Specifically, our method present two key AI innovations. Firstly, we propose a pretext task to extract cross-branch information, which can improve the classification ability by reinforce the consistent predictions between the classification branch and the pretext task. Besides, we propose a pseudo label reassignment approach to adjust the soft classification pseudo labels, and thus suppress pseudo label noise and improve the detection performance. Extensive experiments on Dataset for Object Detection in Aerial Images (DOTAv1.0) and DOTAv1.5 datasets validate the effectiveness and superiority of our method, and demonstrate the practical engineering impact of our method on RS applications and interpretation systems.}, archive = {J_EAAI}, author = {Shitian He and Huanxin Zou and Yingqian Wang and Xu Cao and Hao Chen and Ning Jing}, doi = {10.1016/j.engappai.2025.112378}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112378}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring cross-branch information for semi-supervised remote sensing object detection}, volume = {162}, year = {2025}, }
• (2025). Lightning risk assessment of offshore wind farms considering tropical cyclone impacts based on an improved gradient boosting algorithm. EAAI, 162, 112376. (www)

  @article{ , abstract = {In recent years, global climate warming has led to a significant increase in both the frequency and intensity of tropical cyclones (TCs). The development of TCs is often accompanied by frequent lightning activities. The risk of lightning strikes to high offshore wind turbines is substantially elevated. This study evaluates the lightning risk faced by offshore wind farms influenced by tropical cyclones. Firstly, TC paths are analyzed in both spatial and temporal dimensions by linking them with lightning data to examine the distribution of TC-related lightning, and the lightning strike characteristics of offshore wind turbines are investigated. Secondly, a Bayesian Optimization (BO)-based eXtreme Gradient Boosting (XGBoost) model for lightning risk assessment is proposed, incorporating characteristics of TC lightning and offshore wind farms as input variables. The proposed BO-XGBoost model outperforms XGBoost, Bidirectional Long Short-Term Memory (Bi-LSTM), Support Vector Machine (SVM) and Neural Network (NN), achieving a precision of 98.9 % and a recall of 98.9 % on the test set. Additionally, SHapley Additive exPlanations (SHAP) value analysis indicates that TC lightning characteristics and offshore wind farm characteristics significantly impact the model output, enhancing the accuracy of the model. The assessment outcomes provide a theoretical basis for future offshore wind farm planning and guidance for lightning protection measures in offshore wind farms.}, archive = {J_EAAI}, author = {Qibin Zhou and Kehan Chen and Xiaoyan Bian and Shangjie Chen and Gaopeng Lu}, doi = {10.1016/j.engappai.2025.112376}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112376}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightning risk assessment of offshore wind farms considering tropical cyclone impacts based on an improved gradient boosting algorithm}, volume = {162}, year = {2025}, }
• (2025). A sampling interval-adaptive transformer for industrial time sequence modeling with heterogeneou s sampling rates in quality prediction. EAAI, 162, 112374. (www)

  @article{ , abstract = {The industrial data sequences frequently exhibit irregular sampling frequencies, which pose a number of difficulties for data analysis and modeling. The traditional dynamic models like Recurrent Neural Network (RNN) and Transformer are difficult to model such data sequences. The main reason is that these models assume that data sampling frequency should be constant. To this end, a Sampling Interval-Adaptive Transformer (SIA-Trans) is proposed in this paper to adaptively model the temporal information for heterogeneous sampling sequences in industrial processes. The SIA-Trans uses the sampling interval and position embedding block to address the problem of unequal time intervals and rectify the temporal correlations in time series. Then, the interval-aware self-attention net is designed for dynamic data relationship modeling, taking the processed data through the self-attention mechanism. Finally, the predicted output is obtained after the point-wise feed-forward layer. The proposed SIA-Trans is validated on a real-world hydrocracking process to predict the content of hydrocarbon mixture with five carbon atoms (C5) hydrocarbons in light naphtha, as well as the final boiling point of jet fuel.}, archive = {J_EAAI}, author = {Zijian Xu and Nuo Xu and Kai Wang and Xiaofeng Yuan and Yalin Wang and Chunhua Yang and Weihua Gui and Shuqiao Cheng and Lingjian Ye}, doi = {10.1016/j.engappai.2025.112374}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112374}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A sampling interval-adaptive transformer for industrial time sequence modeling with heterogeneou s sampling rates in quality prediction}, volume = {162}, year = {2025}, }
• (2025). Multi-modality complementary learning network with cross-modality interaction and adaptive fusion for face forgery detection. EAAI, 162, 112373. (www)

  @article{ , abstract = {With the advancement of face forgery technology, many forged faces threaten information security. Although existing face forgery detection methods obtain better detection performance on intra-dataset evaluation, the generalization of cross-dataset detection and the robustness against image post-processing operations still need to be improved. To address these issues, we propose a multi-modality complementary learning network with cross-modality interaction and adaptive fusion for face forgery detection. Specifically, a two-branch architecture is designed to extract spatial and frequency features. To realize the interaction and communication of spatial and frequency information, the cross-modality interaction module is designed to explore the inter-modality correlation by applying across self-attention. Subsequently, a multi-scale feature enhancement module is introduced in the spatial branch to enhance the texture and semantic information of spatial features, improving the robustness of tackling image post-processing operations. In addition, to exploit the complementary relationship between the spatial and frequency features, an adaptive fusion module is designed to establish forged feature dependencies by leveraging spatial self-attention, while learning discriminative feature representations by fusing spatial and frequency features in an adaptive weighted manner. Extensive experimental results on four public datasets demonstrate that the proposed method outperforms other state-of-the-art methods for intra-dataset, cross-dataset, and perturbed dataset evaluations.}, archive = {J_EAAI}, author = {Chunyin Shi and Chengyou Wang and Xiao Zhou and Zhiliang Qin}, doi = {10.1016/j.engappai.2025.112373}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112373}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modality complementary learning network with cross-modality interaction and adaptive fusion for face forgery detection}, volume = {162}, year = {2025}, }
• (2025). Dynamic correlation graph convolution network with embedded temporal correlation extraction for stock price forecasting. EAAI, 162, 112370. (www)

  @article{ , abstract = {Stock price forecasting has become a significant and complex research area within financial technology. The dynamic correlations among stocks and the inherent noise in price volatility present considerable challenges in accurately forecasting stock prices and enhancing investment returns. This paper introduces a novel Dynamic Correlation Graph Convolution Network (DyCGCN) with embedded temporal correlation extraction. First, we propose a dual-scale dynamic graph generation method to capture the topological relationships among stocks. Second, we develop a dynamic correlation-temporal convolution module that extracts high-level temporal correlations. Third, we introduce a prospect theory-guided multi-strategy loss function that accommodates the diverse risk preferences of investors. Furthermore, we present a joint regression-classification learning method to extract and leverage stock trend information. Experiments conducted on four real-world datasets demonstrate the superiority of DyCGCN, achieving an average 24.7% reduction in prediction error and a 10.5% improvement in predictive accuracy over baseline models, underscoring its strong potential for practical stock price forecasting.}, archive = {J_EAAI}, author = {Fang He and Wei Yin and Yilun Jin and Zhengyang Chen}, doi = {10.1016/j.engappai.2025.112370}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112370}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic correlation graph convolution network with embedded temporal correlation extraction for stock price forecasting}, volume = {162}, year = {2025}, }
• (2025). Multi-objective deployment optimization for integrated sensing and communication-enabled unmanned aerial vehicle swarm. EAAI, 162, 112368. (www)

  @article{ , abstract = {With the convergence of mobile communication, sensing, and computational networks in sixth-generation technology, the integration of sensing and communication with unmanned aerial vehicles (UAVs) is promising. This paper focuses on the contribution of artificial intelligence in optimizing the deployment of UAV swarms for multi-objective target detection applications in sixth-generation networks. Specifically, the artificial intelligence contribution lies in the development of an improved multi-objective particle swarm optimization (IMOPSO) algorithm for solving a complex multi-objective deployment problem. The problem aims to simultaneously optimize communication rate, sensing quality, and energy consumption in the deployment of UAV swarms. To address this, the proposed IMOPSO incorporates chaotic initialization, Lévy flight mutation, dynamic mutation rate, and an elimination mechanism based on opposition-based learning. These innovations are designed to enhance the algorithm’s ability to explore the solution space effectively, overcome premature convergence to local solutions, and improve solution quality. In terms of engineering applications, the IMOPSO is applied to the deployment of UAV swarms for target detection, demonstrating its ability to enhance communication and sensing performance while reducing energy consumption in practical scenarios. Through extensive simulations, we show that the IMOPSO outperforms traditional optimization methods and other baseline algorithms, achieving superior results across all optimization objectives. Specifically, the IMOPSO achieves approximately 5% higher transmission data rate, 9% better sensing quality, and 19% lower energy consumption compared to baseline algorithms across multiple test scenarios. Furthermore, the solutions obtained are not only closer to the optimal front but also more concentrated, indicating higher-quality results.}, archive = {J_EAAI}, author = {Hongjuan Li and Haiyuan Chen and Miao Wang and Jiahui Li and Hui Kang and Yuzhuo Guan and Xu Lin}, doi = {10.1016/j.engappai.2025.112368}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112368}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective deployment optimization for integrated sensing and communication-enabled unmanned aerial vehicle swarm}, volume = {162}, year = {2025}, }
• (2025). Progressive semi-supervised learning for specific emitter identification: An iterative clustering and pseudo-label refinement approach. EAAI, 162, 112361. (www)

  @article{ , abstract = {Specific Emitter Identification (SEI) distinguishes radio-frequency (RF) devices by exploiting hardware-induced signal fingerprints, thereby strengthening wireless-layer security. Existing deep learning-based SEI methods depend heavily on labeled data and fixed confidence thresholds for pseudo-labeling, which limits their effectiveness under label scarcity or open-set conditions. To overcome these issues, we propose a progressive semi-supervised learning (ProSSL) method for SEI that combines iterative clustering with contrastive learning to generate adaptive pseudo-labels. ProSSL introduces an “uncertain” class and employs a dual-constraint selector—prediction stability and class diversity—to suppress noisy pseudo-labels and ensure robust propagation. Experiments on the public real-world long range(LoRa) RF-fingerprint dataset show that ProSSL gains 2.90%–6.01% absolute accuracy over state-of-the-art baselines, reaching 96.48% accuracy with 90% labels and 59.88% with only 5% labels. While on the public automatic dependent surveillance-broadcast(ADS-B) Top-10 dataset, ProSSL achieves 84.40% accuracy with 5% labeled data and 99.40% accuracy with 90%labeled data, again outperforming all competing methods. Open-set evaluations further demonstrate that overall accuracy rises from 18.81% when only two classes are known to 62.25% when eight classes are known, confirming strong generalization to unseen emitters and validating ProSSL’s robustness and practicality in realistic wireless environments.}, archive = {J_EAAI}, author = {Yiting Gao and Ke Wang and Hao Huang and Jiao Wang and Jiaxu Liu and Yao Zheng and Jianqing Li}, doi = {10.1016/j.engappai.2025.112361}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112361}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Progressive semi-supervised learning for specific emitter identification: An iterative clustering and pseudo-label refinement approach}, volume = {162}, year = {2025}, }
• (2025). Enhancing scene text image super-resolution via gradient-based graph attention network. EAAI, 162, 112360. (www)

  @article{ , abstract = {Scene text image super-resolution is crucial for enhancing text recognition in low-resolution real-world images. Existing methods usually overlook the structured and repetitive layout of text, which can serve as powerful prior knowledge for guiding reconstruction. In this work, we propose a novel framework that incorporates gradient-based graph attention to explicitly model patch-level text layout. The architecture combines a non-local group-wise attention module, a cascaded channel attention module, and a gradient-guided graph attention module to capture both global and local structural dependencies. This design enables more accurate restoration of text contours and layout consistency. Extensive experiments on the benchmark dataset demonstrate that our method achieves superior performance in both image quality and recognition accuracy, outperforming state-of-the-art methods. The code is available at: https://github.com/cvzxy/TSANv2 .}, archive = {J_EAAI}, author = {Xiangyuan Zhu and Xuchong Liu and Kehua Guo and Wei Zhao}, doi = {10.1016/j.engappai.2025.112360}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112360}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing scene text image super-resolution via gradient-based graph attention network}, volume = {162}, year = {2025}, }
• (2025). A consistency regularization-based approach integrating anatomical structural relationships and organ category representations for multi-organ segmentation in pigs. EAAI, 162, 112359. (www)

  @article{ , abstract = {In modern biomedical research and livestock management, accurate multi-organ segmentation in pigs is essential for breeding programs. However, current methods face challenges due to low imaging contrast, size disparities, and organ shape variability. Additionally, the manual annotation of computed tomography (CT) scans is labor-intensive and costly, limiting available labeled samples. To address these issues, we propose a consistency regularization-based network guided by anatomical structural relationships and global organ category representations, specifically designed for multi-organ segmentation using a limited number of annotated CT scan samples from pigs. Specifically, we designed the SpatialLink Gated Recurrent Unit (GRU) module to extract anatomical structural information and capture dynamic spatial relationships between organs, thereby minimizing segmentation biases caused by organ shape variations. Moreover, we developed the Organ Category Coding module and Guidance module, which integrate consistency regularization and attention mechanisms, enabling the network to accurately extract global organ category representations during the decoding phase, even with a small number of labeled samples, significantly improving segmentation consistency across organs of different sizes. Additionally, We are the first to apply the Visual State Space block to multi-organ segmentation in pigs, using it to extract contextual information. Experiments on 60 pigs demonstrate that our method achieves state-of-the-art results, with significant improvements in segmentation accuracy for the gallbladder and bladder, including a 9.8% and 4.2% Dice score increase, respectively, and a 12.4% and 6.2% boost in Jaccard scores compared to compared with a selection of published methods.}, archive = {J_EAAI}, author = {Xiang Pan and Hang Fan and Jianlan Wang and Yan Fu and Wei Chu and Weipeng Tai and Jing Gu and Jianming Ni}, doi = {10.1016/j.engappai.2025.112359}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112359}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A consistency regularization-based approach integrating anatomical structural relationships and organ category representations for multi-organ segmentation in pigs}, volume = {162}, year = {2025}, }
• (2025). Fast simulation for scattering muography applications using generative adversarial neural networks. EAAI, 162, 112357. (www)

  @article{ , abstract = {Muography is an emergent non-destructive testing technique that uses cosmic muons to probe the interior of objects and structures. This technique can be employed to perform preventive maintenance of critical equipment in the industry in order to test the structural integrity of the facility. Several muography imaging algorithms based on machine learning methods are being developed in the recent years. These algorithms make exhaustive use of simulated data, usually using packages such as GEANT4 (GEometry ANd Tracking), that exhaustively simulate the detector, to produce training samples. This work presents a faster alternative for the generation of simulated samples based on generative adversarial neural networks. A speed up factor of 80 is observed with this system without any significant degradation of the quality of the simulation.}, archive = {J_EAAI}, author = {Rubén López Ruiz and Celia Fernández Madrazo and Sergio Sánchez Cruz and Lara Lloret Iglesias and Pablo Martínez Ruiz del Árbol}, doi = {10.1016/j.engappai.2025.112357}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112357}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast simulation for scattering muography applications using generative adversarial neural networks}, volume = {162}, year = {2025}, }
• (2025). Research on lightweight network for real-time detection of steel structure corrosion based on knowledge distillation. EAAI, 162, 112356. (www)

  @article{ , abstract = {Regular and real-time monitoring of corrosion is crucial for ensuring structural safety and extending the service life of infrastructure. With the continuous development of advanced structural health monitoring technologies, intelligent corrosion detection has become an inevitable trend. This study addresses the issues of low accuracy, incomplete detail processing, and missed or false detections in complex scenarios in steel structure corrosion detection, as well as the challenges of high complexity and insufficient real-time performance in deep learning models. We propose a high-performance, lightweight, real-time corrosion detection model, Real-Time Detection Transformer for Corrosion (RT-DETR-Corrosion), based on knowledge distillation. By incorporating lightweight optimization on the RT-DETR-R18 baseline model and a hybrid knowledge distillation approach, the model significantly improves real-time performance and detection accuracy, meeting the application requirements for efficiency and precision in steel structure corrosion detection. Experimental results show that the model exhibits excellent optimization effects in terms of localization accuracy, classification accuracy, and position regression error on both the training and validation sets, while also demonstrating strong generalization ability. In extreme weather conditions (such as rain, fog, snow, and strong light) and complex scenarios (such as occlusion, blur, and low-light environments), the model maintains stable Precision, Recall, and mAP metrics, validating its reliability and applicability in diverse real-world engineering environments. Moreover, visualized heatmap analysis of detection results for different scenarios further confirms the model's precise attention to corrosion regions and its generalization ability, providing essential technical support for steel structure corrosion risk assessment and intelligent monitoring, with significant potential for engineering applications.}, archive = {J_EAAI}, author = {Jia Hou and Wei Chen and Zhen Duan and Hang Li and Mingyu Yu}, doi = {10.1016/j.engappai.2025.112356}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112356}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on lightweight network for real-time detection of steel structure corrosion based on knowledge distillation}, volume = {162}, year = {2025}, }
• (2025). Function structures of information measures for n-polygonal interval-valued intuitionistic fuzzy sets and their application in location selection strategy. EAAI, 162, 112355. (www)

  @article{ , abstract = {The interval-valued intuitionistic fuzzy set (IVIFS) represents an organic integration of interval-valued fuzzy sets and intuitionistic fuzzy sets, but it fails to meet the requirements of linearity and closure in arithmetic operations, rendering its computations relatively intricate. Consequently, this paper employs an interval partitioning strategy to mitigate the complexity of arithmetic operations, thereby constructing a novel fuzzy set structure characterized by efficient piecewise linear approximation capabilities. Furthermore, the paper presents the structural form of arithmetic operations for the IVIFS with piecewise linear approximation and demonstrates the simplicity of these operations by an numerical example. In addition, we extend the findings on information measures for polygonal interval-valued intuitionistic fuzzy set pertaining to abstract functions that fulfill particular criteria. Furthermore, we delve into the transformation relationships among these information measures, from which a range of structural formats for information measures can be deduced based on functional expressions and transformation relationships. Finally, similarity measures are applied to company site selection. The validity, practicality and stability of proposed measures have been proven through sensitivity analysis and comparative analysis.}, archive = {J_EAAI}, author = {Le Fu and Chunfeng Suo}, doi = {10.1016/j.engappai.2025.112355}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112355}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Function structures of information measures for n-polygonal interval-valued intuitionistic fuzzy sets and their application in location selection strategy}, volume = {162}, year = {2025}, }
• (2025). Two-dimensional refined composite multi-scale revised ensemble dispersion entropy and its application to fault diagnosis of rolling bearing. EAAI, 162, 112354. (www)

  @article{ , abstract = {One-dimensional ensemble dispersion entropy (EDE 1D ) can effectively characterize the nonlinear dynamic characteristics of one-dimensional time series, but the complexity of two-dimensional space is not reflected, and only single-scale features can be captured. Firstly, to comprehensively capture the feature information of two-dimensional space, the symmetrized dot pattern (SDP) is introduced to overcome the shortcomings of the ordinary images lack of physical meaning and the time-frequency distribution methods exhibit incomplete information representation, etc. Simultaneously, the amplitude and frequency information are intuitively expressed by a two-dimensional mirror snowflake symmetrized image (MSSI 2D ). Secondly, to overcome the shortcomings of single-scale and traditional coarse-graining, a two-dimensional refined composite multi-scale coarse-graining method is proposed, which improves the accuracy of feature extraction and reduces the calculation deviation. After that, a new feature extraction method namely two-dimensional refined composite multi-scale revised ensemble dispersion entropy (RCMREDE 2D ) is proposed, whose parameter stability and performance are explored through simulation analysis. The results demonstrate that the RCMREDE 2D exhibits excellent stability and anti-noise interference ability. Based on the advantages of RCMREDE 2D , a novel fault diagnosis method for rolling bearings is developed by integrating RCMREDE 2D and a firefly algorithm optimized support vector machine (FA-SVM) multi-fault classifier for pattern recognition. The proposed method is further validated through two measured bearing data sets and five comparative methods, and the results indicate that the RCMREDE 2D and FA-SVM achieve the highest recognition accuracy while demonstrating superior stability.}, archive = {J_EAAI}, author = {Wenqing Ding and Jinde Zheng and Haiyang Pan and Jian Cheng and Jinyu Tong}, doi = {10.1016/j.engappai.2025.112354}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112354}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-dimensional refined composite multi-scale revised ensemble dispersion entropy and its application to fault diagnosis of rolling bearing}, volume = {162}, year = {2025}, }
• (2025). A review of speaker verification: Methods, network architectures, tasks and challenges. EAAI, 162, 112351. (www)

  @article{ , abstract = {Speaker verification is an important branch of biometric recognition, with wide applications in identity authentication, audio monitoring, and other fields. In recent years, deep learning and meta-learning have made remarkable advancements in the field of speaker verification. Therefore, it is necessary to update existing reviews of speaker verification to reflect the latest research developments. We review literature from the past decade to provide a timely and comprehensive survey of the field. First, we outline the concept and system process of speaker verification. Then, we analyze the speech preprocessing process and common acoustic features used in the systems. Next, we present an overview of speaker modeling approaches, covering traditional probabilistic methods, deep learning-based speaker methods, and meta-learning-based speaker methods, focusing on the latter two methods. We provide an in-depth analysis and summary of the characteristics and the latest network architectures of these methods, focusing on the development of Transformer and large-scale pre-trained Transformer. Furthermore, we introduce the datasets and evaluation metrics used in speaker verification systems, focusing on a detailed and fair comparison of the performance of text-dependent and text-independent speaker verification systems. Finally, we explore the challenges faced by speaker verification systems and discuss future research opportunities.}, archive = {J_EAAI}, author = {Weijie Wang and Hong Zhao and Yikun Yang and Yongjuan Yang}, doi = {10.1016/j.engappai.2025.112351}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112351}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A review of speaker verification: Methods, network architectures, tasks and challenges}, volume = {162}, year = {2025}, }
• (2025). Low-cost and sparsity for continual semantic segmentation. EAAI, 162, 112350. (www)

  @article{ , abstract = {Deep neural networks have contributed to significant progress in semantic segmentation tasks. However, deep neural networks exhibit a critical drop in performance due to catastrophic forgetting when they are required to learn new tasks incrementally. The more plastic the network is, the easier it can learn new tasks. Whereas, for continual semantic segmentation, it is more reliable to preserve the knowledge it has learned from previous tasks. Here, gated 0-1 Bernoulli variable is used as a regularization method to optimize performance by enhancing network sparsity. Then, the special case of gated 0-1 Bernoulli variable is applied in the replay-based method of continual semantic segmentation. Specifically, when the value of the sub-network sampling rate reaches 0.5, the network reaches the strongest stability. Finally, the gated 0-1 Bernoulli variable improves the network’s performance in complex scenarios and reduces cost under similar performance. Experimental results indicate that in using 100% samples for incremental training, the Mean Intersection over Union(mIoU) of the old classes improves by up to 4.6% and 5.5% compared to the baseline at the end of the overall training in continual semantic segmentation scenarios 10-1 and 10-2. Furthermore, in using 60% samples for incremental training, the performance for the old tasks only drops by less than a percentage, while the time cost to complete the full setup decreases by 22%.}, archive = {J_EAAI}, author = {Qing Ji and Bin Li and Shaobo Li and Hongchao An and Jing Yang}, doi = {10.1016/j.engappai.2025.112350}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112350}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-cost and sparsity for continual semantic segmentation}, volume = {162}, year = {2025}, }
• (2025). Promoting sustainability-oriented brand activist campaigns: A spherical fuzzy decision support framework for evaluating activist advertising videos. EAAI, 162, 112349. (www)

  @article{ , abstract = {Activist video advertisements represent a strategic form of brand communication in which companies express their stance on social or environmental issues through emotionally driven storytelling and slogan-based narratives. Despite their growing prevalence, there is a notable lack of systematic and quantitative methods for evaluating their performance or comparing their effectiveness across competing brands. This research addresses that gap by proposing a comprehensive decision support system (DSS) designed to assess the performance of activist video advertisements in a structured and reproducible manner. The study introduces a novel hybrid multi-criteria decision-making (MCDM) framework: the spherical cubic fuzzy (SCF)–Aczel-Alsina–ranking comparison (RANCOM)–method based on the removal effects of criteria (MEREC)–deviation-based pairwise assessment ratio technique (DEPART). This methodology integrates subjective weights obtained via SCF–RANCOM and objective weights derived through SCF–MEREC, with both sets of weights combined using SCF-based aggregation operators that incorporate Aczel-Alsina t-norm and t-conorm functions. Performance rankings are then generated using the SCF–DEPART method. To demonstrate the model's applicability, a real-world case study involving eight sustainability-oriented activist video advertisements released in Türkiye was conducted. Evaluations were based on input from ten domain experts across eleven criteria. The analysis identified “convincingness and credibility” as the most critical factor, with “The Voice of Nature” campaign achieving the highest performance rating. The model's robustness was confirmed through scenario-based sensitivity analyses, and its consistency was validated by benchmarking against thirteen alternative MCDM approaches. The findings offer meaningful implications for both academic research and advertising practice.}, archive = {J_EAAI}, author = {Galip Cihan Yalçın and Karahan Kara and Gülcan Işık and Esra Serdar Tekeli and Vladimir Simic and Abdullah Ballı and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.112349}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112349}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Promoting sustainability-oriented brand activist campaigns: A spherical fuzzy decision support framework for evaluating activist advertising videos}, volume = {162}, year = {2025}, }
• (2025). Federated clustering with mutual knowledge distillation for traffic flow prediction. EAAI, 162, 112347. (www)

  @article{ , abstract = {Traffic flow prediction plays a critical role in intelligent transportation systems. Conventional traffic flow prediction methods primarily rely on centralized training, which poses a risk of privacy leakage. Federated learning, a privacy-preserving framework to machine learning, enables distributed participants to jointly train a shared model without sharing local private data. However, traffic flow is typically collected from different devices and contains different temporal patterns, leading to non-independent and identically distributed. To address these challenges, we propose a traffic flow prediction method based on federated clustering with mutual knowledge distillation. We first perform temporal decomposition on the traffic flow data and use mutual learning with adaptive distillation loss to facilitate mutual knowledge transfer among local models during training. Then, we apply spectral clustering to cluster clients based on the cosine similarity of model parameters at the server and design a global model aggregation method to improve the performance of federated learning. Finally, the proposed method is evaluated on two real-world traffic datasets, and the experiment results show significant improvements over traditional federated learning approaches and also outperform federated mutual learning. The results demonstrate that the proposed method effectively captures temporal information and mitigates the effect of non-independent and identically distributed issues.}, archive = {J_EAAI}, author = {Yao Lin and Shengwu Xiong}, doi = {10.1016/j.engappai.2025.112347}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112347}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated clustering with mutual knowledge distillation for traffic flow prediction}, volume = {162}, year = {2025}, }
• (2025). Quantifying hybrid failure mode in cyclic-loaded reinforced concrete shear walls: Integrating unsupervised and supervised learning techniques. EAAI, 162, 112346. (www)

  @article{ , abstract = {This paper aims to identify failure types in Reinforced Concrete Shear Walls (RCSWs) by quantifying the contributions of shear and flexural modes in the force-deformation response of the walls. The supporting research database includes images, cyclic curve backbones, and geometric and mechanical characteristics of 253 RCSWs. Initially, the database is manually classified into shear, flexure, and shear-flexure failure modes based on observations of surface damage and the cyclic response of the wall. Subsequently, unsupervised clustering and supervised learning algorithms are employed to probabilistically quantify and predict the participation of flexural and shear modes in the overall seismic response of the walls, respectively. The unsupervised model, utilizing the K-means algorithm, identifies the primary failure modes of the walls, achieving over 90 % concordance with manual expert labeling. Based on the results of unsupervised clustering, a hybridity index is proposed to demonstrate the contributions of shear and flexure failure modes to the overall seismic response. Supervised learning is then used to predict hybridity indices from wall characteristics, with the Extremely Randomized Trees (Extra Trees) model achieving the best results based on a balanced evaluation of multiple performance metrics. SHapley Additive exPlanations (SHAP), a tool for exploring model sensitivity, highlights the aspect ratio as the key influencing factor on failure mode, in accordance with relevant structural engineering codes and standards. Implementation of the proposed framework in exploring the behaviors of four unseen case studies reveals a significant correlation between the predicted hybridity indices and observed damage, consistent with existing guidelines.}, archive = {J_EAAI}, author = {Pouya Ebrahimi and Amir Hossein Asjodi and Kiarash M. Dolatshahi}, doi = {10.1016/j.engappai.2025.112346}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112346}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantifying hybrid failure mode in cyclic-loaded reinforced concrete shear walls: Integrating unsupervised and supervised learning techniques}, volume = {162}, year = {2025}, }
• (2025). Dual-path multiple attention-guided feature interaction network for camouflaged object detection. EAAI, 162, 112345. (www)

  @article{ , abstract = {Existing Camouflaged Object Detection (COD) methods rely on features from a pre-trained backbone with a single-encoder structure, leading to initial features biased towards global or local preferences. This affects subsequent feature modeling and degrades COD performance. Some studies use modular or auxiliary flow structures to balance feature preferences but often focus only on interactions between hierarchical features or information flows, ignoring potential redundancy or noise within aggregated features. Therefore, we propose a novel dual-path multiple attention-guided feature interaction network (DMAFI-Net) for COD, which contains four main components: global and local features interaction module (GLFI), intra-feature interaction module (IFI), multi-scale feature enhancement module (MFE), and two decoders including neighbor connection decoder based feature aggregation (NFA) and refine decoder. Specifically, the GLFI is designed to implement the interaction and combination of global and local features, and the combined features will be sent to IFI to mine intra-feature information. Besides, the MFE is introduced to further enrich the extracted features obtained in the IFI. In the feature decoding stage, the NFA module utilizes neighbor connection decoder to fuse multi-scale features and ultimately generates a coarse prediction. Finally, the refine decoder leverages multiple attention modules to refine the initial prediction with the combined features as auxiliary cues and obtain the final camouflaged map. Extensive experiments on four COD benchmark datasets demonstrate the superiority of the proposed framework when compared to 24 state-of-the-art (SOTA) methods in terms of five widely used evaluation metrics. Furthermore, the ablation studies show the effectiveness of main components of our DMAFI-Net.}, archive = {J_EAAI}, author = {Anzhi Wang and Jintao Wu and Shuang Zhao and Yun Liu}, doi = {10.1016/j.engappai.2025.112345}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112345}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-path multiple attention-guided feature interaction network for camouflaged object detection}, volume = {162}, year = {2025}, }
• (2025). TFCTL: Time–Frequency calibrated transfer learning for cross domain depression detection. EAAI, 162, 112344. (www)

  @article{ , abstract = {As the application of speech-based depression detection systems expands, differences in cross-domain data distribution pose significant challenges. This paper proposes the framework of Time–Frequency Calibrated Transfer Learning (TFCTL). This framework first introduces Frequency-Delay Neural Network (FDNN), inspired by Time-Delay Neural Network (TDNN), extending the concept of temporal feature extraction using sliding windows and weight sharing from the time domain to the frequency domain. A multi-level information aggregation module then integrates features of varying abstraction levels from both time-delay and frequency-delay neural networks, balancing global and local speech information. Finally, TFCTL uses transfer learning to calibrate the distribution of the aggregated time–frequency embedding vectors, uncovering commonalities of depression features across different domains. Cross-speaker and cross-corpus experiments were conducted using the Chinese Multimodal Depression Corpus (CMDC) and the Distress Analysis Interview Corpus Wizard-of-Oz (DAIC-WOZ). In cross-speaker scenarios, TFCTL achieved F1 scores of 0.7324 on DAIC and 0.9660 on CMDC, outperforming other methods. In cross-corpus scenarios, TFCTL achieved F1 scores of 0.6743 on DAIC and 0.6879 on CMDC, demonstrating its robustness in addressing domain mismatch issues. The source code used in the paper is available at https://anonymous.4open.science/r/TFCTL-A545/ .}, archive = {J_EAAI}, author = {Dongdong Li and Li Ding and Zuo Yang and Zhe Wang and Ke Zhao}, doi = {10.1016/j.engappai.2025.112344}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112344}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TFCTL: Time–Frequency calibrated transfer learning for cross domain depression detection}, volume = {162}, year = {2025}, }
• (2025). Long-tailed detection and classification of wafer defects from scanning electron microscope images robust to diverse image backgrounds and defect scales. EAAI, 162, 112342. (www)

  @article{ , abstract = {In semiconductor engineering, high yield of wafers relies on accurate detection and classification of wafer defects. The dataset for detecting wafer defects presents three primary challenges: (i) different background types, (ii) variable image or defect scales, and (iii) imbalanced data with a long-tailed distribution of defect types. These challenges create significant limitations for traditional classification techniques. To address these issues, we propose a stratified framework called Wafer Detection and Classification (WaferDC), designed specifically for detecting and classifying wafer defects from scanning electron microscope (SEM) images. Our framework achieves high defect detection performance on SEM wafer images by utilizing a multi-cluster memory bank, which effectively handles the challenges of (i) variable background types and (ii) differing image or defect scales. Building on this robust detection, we propose Segmentation and Mix (SegMix), a novel defect augmentation technique based on anomaly heatmaps, which enhances the reliability of defect detection and classification in a (iii) long-tailed imbalanced environment. Finally, we pass defect-classified images through a parameter-efficient fine-tuning (PEFT)-based classifier (Shiet al., 2023) utilizing a vision transformer (ViT) architecture, further improving overall defect detection and classification performance. We rigorously tested WaferDC on a proprietary SEM wafer dataset and the public Describable Textures Dataset-Synthetic (DTD-Synthetic) and Magnetic Tile Defect (MTD) datasets. The results confirm the effectiveness of our method in improving defect detection and classification in wafer manufacturing. Our code is available at https://github.com/SpatialAILab/WaferDC .}, archive = {J_EAAI}, author = {Taekyeong Park and Yongho Son and Sanghyuk Moon and Seungju Han and Je Hyeong Hong}, doi = {10.1016/j.engappai.2025.112342}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112342}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long-tailed detection and classification of wafer defects from scanning electron microscope images robust to diverse image backgrounds and defect scales}, volume = {162}, year = {2025}, }
• (2025). Fast trajectory planner with a reinforcement learning-based controller for robotic manipulators. EAAI, 162, 112341. (www)

  @article{ , abstract = {Generating obstacle-free trajectories for robotic manipulators in unstructured and cluttered environments remains a significant challenge. Existing motion planning methods often require additional computational effort to generate the final trajectory by solving kinematic or dynamic equations. This paper highlights the strong potential of model-free reinforcement learning methods over model-based approaches for obstacle-free trajectory planning in joint space. We propose a fast trajectory planning system for manipulators that combines vision-based path planning in task space with reinforcement learning-based obstacle avoidance in joint space. We divide the framework into two key components. The first introduces an innovative vision-based trajectory planner in task space, leveraging the large-scale fast segment anything (FSA) model in conjunction with basis spline (B-spline)-optimized kinodynamic path searching. The second component enhances the proximal policy optimization (PPO) algorithm by integrating action ensembles (AE) and policy feedback (PF), which greatly improve precision and stability in goal-reaching and obstacle avoidance within joint space. These proximal policy optimization (PPO) enhancements increase the algorithm’s adaptability across diverse robotic tasks, ensuring consistent execution of commands from the first component by the manipulator, while also enhancing both obstacle avoidance efficiency and reaching accuracy. The experimental results demonstrated the effectiveness of proximal policy optimization (PPO) enhancements, as well as simulation-to-simulation (Sim-to-Sim) and simulation-to-reality (Sim-to-Real) transfer, in improving model robustness and planner efficiency in complex scenarios. These enhancements allowed the robot to perform obstacle avoidance and real-time trajectory planning in obstructed environments. https://sites.google.com/view/ftp4rm/home}, archive = {J_EAAI}, author = {Yongliang Wang and Hamidreza Kasaei}, doi = {10.1016/j.engappai.2025.112341}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112341}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast trajectory planner with a reinforcement learning-based controller for robotic manipulators}, volume = {162}, year = {2025}, }
• (2025). Bi-level location-routing problem with time windows for mixed-load recycling of heterogeneous batteries: A transformer-based improved deep reinforcement learning algorithm. EAAI, 162, 112340. (www)

  @article{ , abstract = {Given the urgency, timeliness and uncertainty of power batteries recycling, we extend a novel network model for retired battery recycling systems, bi-level capacitated location-routing problem with time windows for heterogeneous battery mixed-load (CLRPTW-HBM). Firstly, the expected interval and linear weighting methods are used to transform and process the lower-level multi-objective function that contains fuzzy variables. Secondly, a Transformer-based improved deep reinforcement learning algorithm (Transformer-IDRL) is proposed. (1) The bi-level CLRPTW-HBM is modeled as Markov decision process, and a policy network model with dual-layer encoder-decoder structure is designed based on Transformer architecture. (2) Randomly generate instance data, and the asynchronous advantage Actor-Critic with adaptive dynamic parameter tuning strategy is employed for training. (3) The action sampling strategy based on roulette reverse selection mechanism, and local search strategy incorporating problem characteristics are introduced to improve solution quality. Finally, extensive experiments are conducted on benchmark datasets and actual cases, and the results demonstrate superior performance of Transformer-IDRL, with an average Gap of 0.22 % and 0.19 %, a 5.30 % reduction in recycling cost, and a 6.06 % reduction in battery exposure risk. These satisfactory results highlight the feasibility and efficiency of the proposed model and method. Additionally, sensitivity analysis of model parameters shows that under different decision-maker preferences and vehicle loading capacity, the sensitivity range of path cost is [5.77 %, 39.85 %] and [16.04 %, 35.54 %], while that of exposure risk is [1.40 %, 3.39 %] and [1.35 %, 5.54 %], indicating that parameter variations significant influence the layout of recycling network and overall path cost. Therefore, decision-makers should flexibly adjust key parameters to balance economic benefits and sustainable development in battery recycling.}, archive = {J_EAAI}, author = {Mengna Zhao and Shiping Chen}, doi = {10.1016/j.engappai.2025.112340}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112340}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bi-level location-routing problem with time windows for mixed-load recycling of heterogeneous batteries: A transformer-based improved deep reinforcement learning algorithm}, volume = {162}, year = {2025}, }
• (2025). Instantaneous power prediction for industrial robots using tree-based machine learning methods. EAAI, 162, 112339. (www)

  @article{ , abstract = {This article presents a tree-based machine learning methodology for instantaneous power prediction designed, tested and validated using data from an articulated industrial robot. The proposed methodology for instantaneous power prediction materializes through a generic system architecture with functionalities consisting of data acquisition, time alignment of data samples, storage, model learning, instantaneous power prediction and integration in time to evaluate energy consumption at robot operation level. This methodology is designed to evaluate offsite energy consumption of robotized workstations for different layouts characterized by relative position of the robot with respect to the serviced and fly-by points. This is important both for offline virtual commissioning of robotized workstations (determine layout) and for online operation for maintenance purposes (determine energy spikes different from normal model). The analyzed operation is the linear motion of the robot Tool Control Point in Cartesian space, characterized by the complexity of the kinematic model: each joint operates in coordinated motion, adjusting its velocity and acceleration continuously to ensure a straight path with constant speed. A custom Internet of things (IoT) device enables synchronized energy and motion data logging for robots, ensuring consistent values for sampled trajectories. Justification for the usage of tree-based methods and experimental results are provided.}, archive = {J_EAAI}, author = {Ionuţ Lenţoiu and Silviu Răileanu and Theodor Borangiu and Mihnea Constantinescu and Octavian Morariu}, doi = {10.1016/j.engappai.2025.112339}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112339}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Instantaneous power prediction for industrial robots using tree-based machine learning methods}, volume = {162}, year = {2025}, }
• (2025). Toward sustainable wastewater treatment: Transformer ensembles and multitask learning for energy consumption and quality management. EAAI, 162, 112338. (www)

  @article{ , abstract = {Wastewater treatment plants (WWTPs) are among the most energy-intensive components of urban infrastructure and bear strict regulatory responsibilities for wastewater quality. These dual challenges, minimizing energy consumption and maintaining environmental compliance, are deeply interrelated and must be managed simultaneously to achieve sustainable plant operation. This study proposes a framework that comprises two customized components. The first component employs a voting ensemble model based on transformer architecture to predict energy consumption. It processes heterogeneous feature domains — including hydraulic, wastewater, and climatic variables — through parallel attention-driven streams. The outputs from these streams are then aggregated using a weighted voting mechanism to produce the final prediction. Second, a multitask Bidirectional Gated Recurrent Unit (Bi-GRU) forecasts wastewater quality indicators concurrently (ammonia, Biochemical Oxygen Demand (BOD), and Chemical Oxygen Demand (COD)), capturing shared temporal dependencies and reducing model complexity. A hybrid preprocessing strategy is applied, incorporating domain-aware outlier detection (z-score and Interquartile Range (IQR)), K-Nearest Neighbors (KNN) Imputation, and feature selection using Extreme Gradient Boosting (XGBoost). Experimental results showed that. The voting ensemble model achieved the best results for energy consumption prediction with 31.61 of Root Mean Squared Error (RMSE). The multitask Bi-GRU achieved the best results for wastewater quality indicators with RMSE at 6.1689, 48.0323, and 88.2214 for ammonia, BOD, and COD, respectively. This work is among the first to integrate transformer ensembles and multitask learning in a unified WWTP forecasting system. Simultaneously addressing energy efficiency and water quality assurance, this offers a practical, scalable, and intelligent decision-support tool for sustainable wastewater management.}, archive = {J_EAAI}, author = {Hager Saleh and Sherif Mostafa and Shaker El-Sappagh and Abdulaziz AlMohimeed and Michael McCann and Saeed Hamood Alsamhi and Niall O’Brolchain and John G. Breslin and Marwa E. Saleh}, doi = {10.1016/j.engappai.2025.112338}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112338}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Toward sustainable wastewater treatment: Transformer ensembles and multitask learning for energy consumption and quality management}, volume = {162}, year = {2025}, }
• (2025). Complete information extraction for monocular depth estimation using a dual framework. EAAI, 162, 112337. (www)

  @article{ , abstract = {This paper aims to address the problem of efficient extraction of complete multi-scale information for supervised monocular depth estimation. Most of the existing depth estimation methods are based on Convolutional Neural Network (CNN). By gradually exploring the contextual and semantic features, they have achieved good results in scene depth estimation. However, with the expansion of the receptive field, global information limited by the local induction bias is gradually suppressed, resulting in the performance cannot be further improved. Recently, Transformer-based methods have been widely used to model the global correlation between features. Nevertheless, since the Transformer networks are not spatially aware enough, they usually lose local details and have no clear mechanism for reusing features when processing images. The Transformer networks perform self-attention mechanism at each location and cannot directly obtain information from other locations for features. Therefore, we propose a novel dual framework called as Transformer-CNN, which includes the Transformer-branch and the CNN-branch for monocular depth estimation. Specifically, the Transformer-branch is able to model the global contextual information and the CNN-branch can capture local spatial relationships in images. However, simply fusing these two independent branches may result in insufficient feature aggregation. To this end, we design a Parallel Feature Interaction Module (PFIM), which contains a Self-Attention Module (SAM) and a Cross-Attention Module (CAM), so as to highlight features from the Transformer-branch and the CNN-branch respectively and extract complementary information between the two branches. Meanwhile, in order to make full use of the low-level features with low quality in the scene, we propose a Low-level Information Acquisition Module (LIAM) to capture texture-related information and preserve texture details in the CNN-branch. Finally, to address the lack of multi-scale contextual information in Vision Transformer (ViT), we introduce a Wide Area Multi-scale Decoder (WAMD), which incorporates the multi-scale feature representations into the decoder part via a Wide Area Attention (WAA). Extensive experiments on benchmark datasets collected in the outdoor and indoor environments demonstrate the competitive results of the proposed method, compared with the state-of-the-art monocular depth estimation methods.}, archive = {J_EAAI}, author = {Bin Li and Dazheng Zhou and Xianjie Gao and Mingliang Zhang}, doi = {10.1016/j.engappai.2025.112337}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112337}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Complete information extraction for monocular depth estimation using a dual framework}, volume = {162}, year = {2025}, }
• (2025). Beyond the horizon: A comprehensive analysis of artificial intelligence-based weather forecasting models. EAAI, 162, 112335. (www)

  @article{ , abstract = {The field of Artificial Intelligence (AI)-based weather forecasting is growing rapidly, with continuous progress in model development, techniques, and performance improvements. This paper provides a comprehensive overview of AI-based weather forecasting models, focusing on their current status, challenges, and directions for further development. A review of more than 40 models, primarily proposed after 2015, underscores the importance of critically examining various aspects of AI-based forecasting. Unlike previous reviews that targeted only a limited number of models or features, this study addresses a complete set of aspects and analyzes existing challenges from multiple perspectives. These aspects include the Machine Learning (ML) and Deep Learning (DL) methods used, datasets, predictand parameters, overfitting, and capability for forecasting extreme weather, lead time, spatiotemporal scale, performance criteria, overfitting, data assimilation, data-driven models, and the analysis of state-of-the-art (SOTA) models such as FengWu, ClimaX, Pangu-Weather, FourCastNet, GraphCast, GenCast, and Artificial Intelligence Forecasting System (AIFS) from various viewpoints. The review also discusses current challenges, including limited historical data and data quality, small-scale weather forecasting, model explainability, uncertainty, extreme weather prediction, physical constraints, temporal adaptation, and generalization, and outlines potential future directions.}, archive = {J_EAAI}, author = {Saeid Haji-Aghajany and Witold Rohm and Piotr Lipinski and Maciej Kryza}, doi = {10.1016/j.engappai.2025.112335}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112335}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Beyond the horizon: A comprehensive analysis of artificial intelligence-based weather forecasting models}, volume = {162}, year = {2025}, }
• (2025). Neural model of the adaptive tuned particle impact damper. EAAI, 162, 112334. (www)

  @article{ , abstract = {The paper presents a novel approach for the modeling of the Adaptive Tuned Particle Impact Damper (ATPID) using Multilayer Perceptron (MLP). The main motivation was the recognition that such an approach can support the development of novel neural modeling and the optimal determination of damper parameters in terms of mechanical vibration attenuation. The training data were obtained using a theoretical model validated experimentally. The optimally selected MLP was compared with other regression models using 10 different metrics. A hyperparameter tuning of the determined neural network architecture was conducted based on the input parameters such as excitation amplitude, grain mass, and ATPID damper height. The analyses show that the proposed neural network could quickly and accurately estimate the system’s vibration amplitude and efficiently predict the optimal damper height. The ability to effectively determine the correct optimal height is crucial for ATPID damper control. The high efficiency in predicting the system’s vibration amplitude allows for the replacement of the theoretical model with applied time-consuming contact forces. The MLP accurately estimated vibration amplitudes with 1%–10% error for interpolated data and up to 15% for extrapolated cases. The issue raised is particularly important from the perspective of real-time damper control. It was found that computing a single case using the artificial neural network is more than ten times faster compared to the theoretical model. Therefore, the proposed ATPID damper model based on a neural network forms the basis for further considerations and scientific research to finally propose a control algorithm in the future.}, archive = {J_EAAI}, author = {Mateusz Żurawski and Karolina Grabska and Robert Zalewski and Adam Kulawik}, doi = {10.1016/j.engappai.2025.112334}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112334}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural model of the adaptive tuned particle impact damper}, volume = {162}, year = {2025}, }
• (2025). An adaptive hybrid machine reading comprehension framework for multimodal emotion-cause pair extraction in conversations. EAAI, 162, 112333. (www)

  @article{ , abstract = {Multimodal emotion-cause pair extraction in conversations (MECPEC) has gradually evolved into an emerging task aimed at discovering deeper causal relationships between emotions and their corresponding causes in conversational contexts. It has widespread application in fields such as human–computer interaction, social media analysis, customer feedback management, and empathetic companion, among others. However, the challenges posed by the oversimplified multimodal feature fusion mechanism and the failure to account for the relative positional relationship between emotions and causes still hinder the performance of MECPEC. In this study, we propose an adaptive hybrid machine reading comprehension (AHMRC) framework to extract potential emotion-cause pairs inherent in conversations. The MECPEC task is first transformed into a two-round hybrid machine reading comprehension task that sequentially enforces the global emotion query and the local cause query with the goal of exploring the relative position constraint specific to conversations. Subsequently, an adaptive multimodal attention module is designed by incorporating features extracted from text, video, and audio modalities, and adaptively fusing them according to their contributions. Extensive experiments were carried out on the benchmark datasets to demonstrate the effectiveness of the proposed AHMRC framework in comparison to other state-of-the-art methods in the literature.}, archive = {J_EAAI}, author = {Guorui Li and Xufeng Duan and Cong Wang and Sancheng Peng}, doi = {10.1016/j.engappai.2025.112333}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112333}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive hybrid machine reading comprehension framework for multimodal emotion-cause pair extraction in conversations}, volume = {162}, year = {2025}, }
• (2025). Analytical and machine learning approaches for three-dimensional orthotropic functionally graded cylindrical structure in steady heat conduction. EAAI, 162, 112332. (www)

  @article{ , abstract = {Theoretical and computational modelling of heat conduction in a functionally graded cylinder has been rigorously investigated in prior studies, owing to its critical significance in high-stakes engineering applications such as nuclear reactor design, aerospace structural systems, and pressure vessel technology. Despite this extensive body of work, the majority of these studies have mostly focused on simplified two-dimensional models, frequently presuming idealized thermal parameters such as isotropic thermal conductivity, constant convection coefficients, and spatially uniform ambient temperatures. These simplifications overlook the crucial role of spatially varying thermal characteristics and three-dimensional (3D) temperature distributions, which are essential for accurately simulating the complex heat conduction behaviors found in real-world engineering. Unlike prior research, this study presents an analytical solution for heat conduction under non-homogeneous generalized Robin boundary conditions, capturing 3D thermal conductivity inhomogeneities along three orthogonal directions using Sturm-Liouville theory and finite integral transforms. However, while such analytical methods are highly accurate for simpler geometries, they often encounter significant challenges when extended to scenarios involving complex material gradients and irregular domains. A gradient-enhanced physics-informed neural network (g-PINN) framework is proposed to tackle these challenges, utilizing neural networks that incorporate physical laws and gradient information to enhance its applicability to complex configurations. This combined approach presents a novel framework that integrates classical theory with machine learning, facilitating precise modelling of thermal phenomena in functionally graded cylinders. The findings indicate that both the analytical and machine learning approaches (g-PINN) align closely with presented solutions, precisely capturing the energy equation and more complex boundary conditions.}, archive = {J_EAAI}, author = {Palash Das and Md Ashraful Islam and Dipayan Mondal}, doi = {10.1016/j.engappai.2025.112332}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112332}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analytical and machine learning approaches for three-dimensional orthotropic functionally graded cylindrical structure in steady heat conduction}, volume = {162}, year = {2025}, }
• (2025). Robust spurious drug–drug interaction detection via chi-square-guided bidirectional attention mechanism. EAAI, 162, 112331. (www)

  @article{ , abstract = {Spurious drug–drug interactions arising from experimental biases can compromise treatment safety and hinder effective clinical decision-making for patients. However, current molecular representation learning methods still face significant challenges in addressing this problem. First, most graph neural network-based approaches consider only single-direction interactive semantic extraction between drug molecules, failing to capture more granular relationships between drugs fully. Moreover, during training, they are vulnerable to noise from negative sampling and cannot sufficiently leverage the complete drug–drug interaction annotations, resulting in reduced robustness and accuracy in downstream tasks. To this end, we propose a robust spurious DDI detection framework that employs chi-square-guided bidirectional attention to capture fine-grained and bidirectional interaction patterns. First, considering their mutual information flow, a two-way cross-attention mechanism is introduced for a more granular extraction of cross-drug molecular interaction semantic representations through bidirectionally perceiving interactive features between drugs. Second, on the basis of robust minimum covariance determinant theory, we propose a chi-square distribution-based spurious detection method to approximate the correctly annotated drug–drug interactive feature space to a chi-square distribution for a complete feature representation. Extensive experiments on benchmark datasets further validate our method’s effectiveness over state-of-the-art methods, particularly in noisy interference scenarios. Our code is available at https://github.com/AlexCostra/cd .}, archive = {J_EAAI}, author = {Wei-Yu Shi and Yi-Jia Zhang and Jin-Zhong Ning}, doi = {10.1016/j.engappai.2025.112331}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112331}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust spurious drug–drug interaction detection via chi-square-guided bidirectional attention mechanism}, volume = {162}, year = {2025}, }
• (2025). A domain adaptation method to defend chinese textual adversarial attacks via prompt-tuning. EAAI, 162, 112330. (www)

  @article{ , abstract = {The textual adversarial attack aims to fool existing models into making erroneous predictions by adding strategic perturbations to normal data without affecting the user’s understanding. Recently, methods based on Pre-trained Language Models (PLMs) and Large Language Models (LLMs) have shown promising performance in various Natural Language Processing (NLP) downstream tasks. However, due to significant deviations between the original and perturbed texts, these methods struggle to achieve satisfactory results in defending against textual adversarial attacks, especially in Chinese, which has unique syntactic structures. To address this issue, we propose a domain adaptation method for defending against Chinese textual adversarial attacks through a prompt-tuning model, which effectively mitigates the discrepancy between different domains. Specifically, the original and perturbed texts are treated as the source and target domains, respectively, with the textual adversarial defense task framed as a cross-domain classification problem. The soft prompt-tuning model trained in the source domain is iteratively adapted to uncover the true label information in the target domain. The graph attention network is incorporated to integrate Chinese syntactic structure information with semantic features. Through a voting mechanism on predicted labels generated by the iterative model, soft prompt-tuning is further optimized for cross-domain classification tasks. Extensive experimental results demonstrate the superior effectiveness of our method in Chinese textual adversarial defense tasks compared to baseline methods, including the state-of-the-art fine-tuning approaches for PLMs and LLMs.}, archive = {J_EAAI}, author = {Yi Zhu and Zhenglong Li and Yun Li and Yunhao Yuan and Jipeng Qiang}, doi = {10.1016/j.engappai.2025.112330}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112330}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A domain adaptation method to defend chinese textual adversarial attacks via prompt-tuning}, volume = {162}, year = {2025}, }
• (2025). A novel high-accuracy graph neural network-based rumor detection method. EAAI, 162, 112329. (www)

  @article{ , abstract = {Rumors spreading on social media platforms result in potential damages. A precise rumor detection mechanism can help form a healthy public opinion environment. In recent years, deep learning-based rumor detection methods, especially graph model-based ones, have risen and reached promising performance. However, there are several defects in existing methods, which limit models from efficiently utilizing the propagation structure. In this paper, we propose a novel rumor detection model, which has high accuracy and reaches state-of-the-art performance. First, we design a powerful comprehensive rumor feature extractor that explicitly overcomes the restriction of previous Graph Neural Networks-based models. Then, by introducing Kernel Subtree features, our model acquires the capability to learn crucial local features from important nodes. Comparative experiments performed on two real-world social media platforms demonstrate that our work reaches state-of-the-art performance, which outperforms the best baseline with 1.6% and 1.9% in accuracy respectively.}, archive = {J_EAAI}, author = {Xi Xiao and Zeming Wu and Chengzong Cai and Tian Bian and Guangwu Hu and Qing Li and Cheng Huang}, doi = {10.1016/j.engappai.2025.112329}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112329}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel high-accuracy graph neural network-based rumor detection method}, volume = {162}, year = {2025}, }
• (2025). Committee of multi-scale nonlinear learning frameworks for accurate stock price forecasting. EAAI, 162, 112325. (www)

  @article{ , abstract = {Dependable stock price predictions are vital for optimizing economic policies and investment strategies in both national and corporate settings. However, the intrinsic volatility and intricacy of stock prices pose considerable challenges. Thus, this paper introduces a novel Committee of Multi-scale Nonlinear Learning Frameworks (CoML) that employs a three-stage model: decomposition, reconstruction, and prediction. First, a complete ensemble empirical mode decomposition with adaptive noise is adopted to decompose the original stock prices into multiple intrinsic mode functions. Secondly, a fine-to-coarse algorithm is applied to reconstruct the intrinsic mode functions, so as to effectively extract short-term fluctuations and long-term trends. Finally, an ensemble of nonlinear models including bidirectional long short-term memory (BiLSTM), support vector regression (SVR) and multi-layer perceptron (MLP) is used to learn and forecast features extracted to obtain high performance. Experimental results indicate that the model performs exceptionally well in both emerging and developed markets highlighting the innovative capabilities of CoML in highly complex and volatile financial markets. The proposed model is further validated using Model Confidence Set and the results indicate that the model is statistically significant.}, archive = {J_EAAI}, author = {Qian He and Yanhui Liang and Yu Lin and Dazhi Pan and Yuying Yue}, doi = {10.1016/j.engappai.2025.112325}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112325}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Committee of multi-scale nonlinear learning frameworks for accurate stock price forecasting}, volume = {162}, year = {2025}, }
• (2025). Learning robust brain tumor segmentation under label corruption and data scarcity. EAAI, 162, 112322. (www)

  @article{ , abstract = {Deep learning models for medical image segmentation often struggle with performance issues when datasets are affected by label noise or annotation errors, commonly introduced during manual process or lack of proficiency. These noisy annotations disrupt the loss function, leading to "partially incorrect" gradients that impair the model's learning and overall performance. Additionally, the limited availability of scanned data for training often makes it challenging to develop a robust model. A common approach to address this issue is to leverage similar large annotated datasets. However, differences in dataset distributions can also lead to inconsistencies, introducing erroneous gradients during training and further impacting model performance. To address these challenges, we propose MGR-DAS (Meta-Gradient Reweighting via Direction-Aware Similarity), a novel meta-learning-based approach that can automatically evaluate the reliability of training samples during training using a small, clean subset easily curated from the noisy dataset. Our method quantifies reliability by measuring the cosine similarity between the gradients of noisy training samples and those of the clean subset. Samples with higher gradient alignment are assigned greater weights during training, effectively reducing the impact of noisy labels and improving model robustness. We evaluate our method using three standard metrics for medical image segmentation: the Dice Similarity Coefficient (DSC), the 95th percentile Hausdorff Distance (HD95), and Intersection over Union (IoU). The proposed MGR-DAS achieved an overall 2.4 % improvement in the DSC on the brain tumor segmentation (BraTS, 2021) dataset. Remarkably, even with only 10 clean annotations used in the reweighting algorithm, our method yielded a 28.7 % gain in DSC. In real-world, data-scarce scenarios, our proposed MGR-DAS also improved the overall DSC score by 2.6 % on BraTS pediatric (BraTS-PEDs) and by 1.0 % on BraTS-Africa, demonstrating strong generalizability and robustness. Experimental results confirm that the proposed method reliably identifies noisy data, prioritizes clean data through adaptive weighting, and outperforms existing fine-tuning, curriculum learning techniques, and other meta-learning frameworks commonly employed in classification tasks.}, archive = {J_EAAI}, author = {Abdulkhalek Al-Fakih and Abbas Mohamed Rezk and Abdullah Shazly and Kanghyun Ryu and Mohammed A. Al-masni}, doi = {10.1016/j.engappai.2025.112322}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112322}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning robust brain tumor segmentation under label corruption and data scarcity}, volume = {162}, year = {2025}, }
• (2025). Zero-shot learning augmented slow feature analysis for semantic-aware industrial process fault detection. EAAI, 162, 112321. (www)

  @article{ , abstract = {Slow Feature Analysis (SFA) has shown considerable success in the field of industrial process fault detection. Nonetheless, due to its unsupervised nature, SFA relies solely on the normal training data and overlooks the incorporation of prior process knowledge, which consequently diminishes its efficacy in early fault detection. To mitigate this limitation, this paper introduces the concept of Zero-Shot Learning (ZSL) and proposes an improved SFA approach, referred to as ZSL-SFA. This novel method leverages fault semantic representations as auxiliary knowledge to enhance fault detection sensitivity in industrial process monitoring. The ZSL-SFA framework implements a dual-model collaborative monitoring system: (1) a primary SFA model is developed using normal operational data to capture the dynamic characteristics of the process; and (2) a semantic encoding mechanism, grounded in expert knowledge, is devised to build the auxiliary model, where a probabilistic attribute learner adaptively extracts semantic information from fault attribute descriptions, facilitating effective fault knowledge transfer through similarity analysis. The monitoring outcomes from both the primary and auxiliary models are integrated using a Bayesian fusion strategy, culminating in a comprehensive ZSL-SFA monitoring system. The main advantage of this method is its ability to fully exploit prior process knowledge to enhance the basic SFA model without the need for additional labeled fault samples. Experimental validations on the Tennessee-Eastman process simulation platform are performed to indicate that the proposed ZSL-SFA method surpasses the basic SFA method in terms of fault detection performance.}, archive = {J_EAAI}, author = {Wenjie Yang and Xiaogang Deng and Lumeng Huang and Yuping Cao}, doi = {10.1016/j.engappai.2025.112321}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112321}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Zero-shot learning augmented slow feature analysis for semantic-aware industrial process fault detection}, volume = {162}, year = {2025}, }
• (2025). Multi-scale embedding with guided attention for medical image analysis. EAAI, 162, 112319. (www)

  @article{ , abstract = {Accurate classification of medical images is pivotal for enhancing diagnostic accuracy and optimizing treatment strategies. Traditional methods encounter challenges in delineating clear inter-class boundaries within high-dimensional feature spaces affected by class overlap. This study introduces a Multi-Scale Embedding with Guided Attention (MSEGA) framework based on deep learning autoencoders that integrates innovative guided attention learning mechanisms without explicit target mask supervision for tumor detection and classification. The framework incorporates Multi-scale Feature Extraction Blocks and Depthwise Separable Convolution Blocks to comprehensively capture image features. Through Channel Attention and Spatial Attention mechanisms, the MSEGA method prioritizes critical tumor regions across scales. We also propose a novel interpretable embedding learning loss function to optimize image embeddings, highlighting crucial regions and refining category distinctions. Empirical evaluations on two brain tumor Magnetic Resonance Imaging (MRI) datasets demonstrate our approach surpasses conventional methods including Convolutional Neural Networks and Vision Transformers in classification accuracy and generalizability. These results underscore the framework’s potential as a versatile artificial intelligence-powered tool in medical image analysis.}, archive = {J_EAAI}, author = {Zeyan Li and Yifei Peng and Yizun Lin}, doi = {10.1016/j.engappai.2025.112319}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112319}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale embedding with guided attention for medical image analysis}, volume = {162}, year = {2025}, }
• (2025). Multimodality based deep learning method for cancer-related T-cell receptor sequence prediction. EAAI, 162, 112318. (www)

  @article{ , abstract = {T-cell receptor sequences (TCR-seq) are closely related to cancers, and in particular, cancer-related TCR-seq are crucial in cancer diagnosis and treatment. Current prediction methods for cancer-related TCR-seq often focus solely on the sequence structure, neglecting its spatial structure. Therefore, we propose a multimodal deep learning method based on parallel and residual structures (MDPR) for the detection of cancer-related TCR-seq. MDPR can effectively integrate the spatial and sequence structure of TCR-seq for accurately identifying cancer-related sequences. First, we introduce a TCR-seq encoding method based on atomic three-dimensional spatial coordinates, allowing for more effective extraction of the spatial structural features of TCR-seq. Second, we use high-dimensional word vectors instead of the amino acid feature vectors traditionally used by other researchers. Third, we pretrain the spatial feature extraction module and then conduct joint training with the sequence feature extraction module. This approach allows the model to better consider the relationship between the two modalities, thereby improving prediction accuracy. Finally, MDPR achieved an area under the curve (AUC) of 0.971 after ten rounds of three-fold cross-validation on the dataset. The AUC of MDPR is 5% higher than that of the previous best method. In short, we propose an artificial intelligence method called MDPR, and apply it to the biomedical field. MDPR can be obtained from https://github.com/biomg/MDPR .}, archive = {J_EAAI}, author = {Junjiang Liu and Shusen Zhou and Mujun Zang and Chanjuan Liu and Tong Liu and Qingjun Wang}, doi = {10.1016/j.engappai.2025.112318}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112318}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodality based deep learning method for cancer-related T-cell receptor sequence prediction}, volume = {162}, year = {2025}, }
• (2025). A combined perspective self-supervised contrastive learning framework for human activity recognition integrating instance prediction and clustering. EAAI, 162, 112317. (www)

  @article{ , abstract = {Human activity recognition (HAR) is a significant application for wearable devices, which primarily identifies current human activities by analyzing sequential sensor data. The real-time data recording of wearable devices enables the collection of vast amount of unlabeled data. Utilizing this data for self-supervised contrastive pre-training of HAR models presents a feasible solution to the decline in recognition performance due to limited labeled data. However, traditional contrastive learning frameworks are primarily designed based on positive and negative sample pairs in the image domain. The relatively simple sequence data of HAR is prone to generating incorrect negative pairs, thus pre-training HAR models solely in this manner is unsatisfactory. Given the phenomena described above, this paper proposes an Instance Prediction and Clustering Self-supervised Contrastive Learning Framework (IPCSC) for HAR, considering the characteristics of human activity data. IPCSC circumvents negative sample pairs, instead extracting contrastive information at the instance perspective by prediction tasks among various augmented views of samples and integrating clustering concepts for contrastive learning from a holistic perspective. The primary objective is to enable the model to discern critical information within human activity data and distinguishable features between different activities, thereby improving the model’s pre-training efficacy and enhancing its downstream activity recognition performance. Numerous experimental analyses demonstrate that IPCSC outperforms other self-supervised methods, achieving an average F1-Score performance improvement of 5.65%, 4.11%, and 7.99% over supervised baselines on the UCI-HAR, MobiAct, and MotionSense datasets, respectively, with only 1% of the labeled data.}, archive = {J_EAAI}, author = {Zhixuan Yang and Kewen Li and Zongchao Huang and Zhifeng Xu and Xinyuan Zhu and Yuan Xiao}, doi = {10.1016/j.engappai.2025.112317}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112317}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A combined perspective self-supervised contrastive learning framework for human activity recognition integrating instance prediction and clustering}, volume = {162}, year = {2025}, }
• (2025). A self-supervised method for learning path-augmented knowledge graph embedding. EAAI, 162, 112315. (www)

  @article{ , abstract = {Knowledge graphs (KGs) consist of factual triples that describe relations between entities in the real world. Knowledge graph embedding (KGE) aims to map entities and relations into constantly low-dimensional vectors, which is important for lots of downstream tasks (e.g., KG completion and information retrieval). Current KGE methods primarily rely on explicit structural patterns, neglecting latent contextual semantics behind those structures and resulting in sub-optimal performance. While some methods incorporate additional data (e.g., textual descriptions), such dependencies limit applicability due to additional data requirements. Furthermore, most KGE models suffer from limited supervision with sparse labeled triples, restricting their capacity to learn comprehensive semantic features. Inspired by the simple but effective self-supervised language model word2vec, one interesting question is: Can KGE be performed as a simple self-supervised language model ? To achieve this, we innovatively propose a self-supervised KGE framework that learns entity and relation embeddings by adapting word2vec’s skip-gram objective to path sequences extracted from KGs. Our framework employs separate embedding spaces for entities and relations with an entity-relation mapping mechanism for effective interaction between the two embedding spaces. Further, to enhance the training efficiency, we introduce a markov chain-based negative sampling strategy, which generates semantically meaningful negative samples by preserving the structural contexts along KG paths. Our framework, which is the first attempt to follow the context-based self-supervised idea of language models to conduct KGE tasks, addresses the constraints of label-dependent supervised KGE techniques and obviates the requirement for external information, while simultaneously enabling effective extraction of the implicit contextual semantics inherent in triple structures. Experiments on two widely-used KGE datasets show state-of-the-art performance, demonstrating our framework’s ability to learn semantically rich representations solely from graph structure.}, archive = {J_EAAI}, author = {Tong Shen and Fu Zhang and Jingwei Cheng}, doi = {10.1016/j.engappai.2025.112315}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112315}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-supervised method for learning path-augmented knowledge graph embedding}, volume = {162}, year = {2025}, }
• (2025). A single-cell RNA sequencing data imputation method based on non-negative matrix factorization and multi-kernel similarity network fusion. EAAI, 162, 112313. (www)

  @article{ , abstract = {Artificial intelligence-based single-cell RNA sequencing (scRNA-seq) technology is widely used in cell type identification and disease research, but its data often contain a large number of missing values and zero values due to technical limitations and biological differences. These zero values not only affect downstream analysis, but also make it difficult to distinguish technical zero values from biological zero values. Therefore, this paper proposes a scRNA-seq data interpolation method (sc-MKNMF) based on non-negative matrix factorization and multi-kernel similarity network fusion for the first time. This method improves the accuracy of cell clustering by accurately filling some zero values. First, sc-MKNMF uses gene-cell dual-level analysis to distinguish technical zero values from biological zero values, and then calculates the similarity network of multi-kernel fusion of genes and cells respectively. Then, this method uses non-negative matrix factorization combined with similarity network to construct the objective function, and introduces sparse regularization terms to ensure the similarity between genes and cells and improve stability. In addition, sc-MKNMF is also equipped with an efficient optimization algorithm to promote its convergence by continuously updating the objective function. Finally, the verification and comparative experiments on 12 scRNA-seq datasets show that the sc-MKNMF method outperforms other advanced data interpolation methods. In addition, the extension of sc-MKNMF to the two tasks of cell trajectory inference and differentially expressed gene analysis showed significant improvement and excellent versatility.}, archive = {J_EAAI}, author = {Pei Liu and Cheng Chen and Hao Liu and Jin Gu and Xinya Chen and Ying Su and Zhiyuan Cheng and Xiaoyi Lv and Chen Chen}, doi = {10.1016/j.engappai.2025.112313}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112313}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A single-cell RNA sequencing data imputation method based on non-negative matrix factorization and multi-kernel similarity network fusion}, volume = {162}, year = {2025}, }
• (2025). A novel object detection model for sugar beet cercospora leaf spot in field scenarios based on large kernel decomposition and spatial channel interaction attention. EAAI, 162, 112311. (www)

  @article{ , abstract = {Cercospora leaf spot (CLS) is a widespread disease that seriously threatens beet yield and sugar quality. Timely detection enables farmers to take early control measures and reduce economic losses. Although artificial intelligence (AI)-based methods are replacing manual inspection in agriculture, CLS detection in complex field environments remains highly challenging due to subtle early-stage symptoms and severe occlusions caused by overlapping leaves and weeds. To address these challenges, this paper presents Cercospora Leaf Spot–You Only Look Once (CLS–YOLO), an enhanced detection model built upon You Only Look Once version 11 (YOLOv11), incorporating novel modules specifically designed for accurate CLS detection under challenging field conditions. To improve the detection of weak and early-stage symptoms, we design the Multi-Scale Large Kernel Decomposition (MSLKD) module, which enhances feature extraction for subtle lesions. Furthermore, we develop the Spatial-Channel Interaction Attention (SCIA) module to mitigate detection errors arising from occlusion and fragmented disease patterns by refining multi-scale feature representations. Experimental results demonstrate CLS–YOLO achieves superior performance, reaching an mAP@0.5 of 73.6% ± 0.2% and an mAP@0.5:0.95 of 40.6% ± 0.3% over five independent runs, outperforming twelve mainstream object detection algorithms while maintaining lightweight efficiency. To validate generalization capability across scenarios, crops, and diseases, we conducted comparative experiments on two public crop disease datasets, where our method achieved superior overall performance. In summary, this study provides an effective AI-driven solution for precise crop disease detection, contributing to the practical advancement of intelligent agriculture.}, archive = {J_EAAI}, author = {Hualong Dong and Yi Lu and Yurong Qian and Xuefei Ning and Ting Chen and Ke Tang}, doi = {10.1016/j.engappai.2025.112311}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112311}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel object detection model for sugar beet cercospora leaf spot in field scenarios based on large kernel decomposition and spatial channel interaction attention}, volume = {162}, year = {2025}, }
• (2025). ElecBench: A large language model benchmark in electric power domain. EAAI, 162, 112310. (www)

  @article{ , abstract = {Large language models (LLMs) have made substantial advancements in the field of natural language processing, necessitating the development of new benchmarks to accurately track their progress. In this paper, we introduce ElecBench, the first benchmark specifically designed for the electric power domain. ElecBench comprises 24 datasets spanning different scenarios, covering general electric power knowledge and four specific business applications, with a total of 34,030 data entries. Furthermore, we evaluate the performance of a series of open-source Chinese LLMs on ElecBench. Our experiments demonstrate that ElecBench serves as an effective benchmark for electric power scenarios and highlight that existing LLMs require further optimization to gain domain-specific knowledge and achieve better performance.}, archive = {J_EAAI}, author = {Sai Zhang and Qiaochu Huang and Qiang Zhang and Xiao Liang and Weiwei Liu and Kunlun Gao and Fei Zhou and Congcong Shi}, doi = {10.1016/j.engappai.2025.112310}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112310}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ElecBench: A large language model benchmark in electric power domain}, volume = {162}, year = {2025}, }
• (2025). Feature-enhanced edge-INverse attention network for skin lesion segmentation. EAAI, 162, 112306. (www)

  @article{ , abstract = {Skin cancer, particularly melanoma, is one of the most aggressive and deadly forms of cancer with its incidence rising globally. Early detection is crucial for improving survival rates, but the traditional dermatoscopy method is a highly time-consuming and subjective process. To resolve this issue, we propose a novel Feature-Enhanced Edge-INverse attention network (FEEINnet) model that helps to segment the skin lesion region more accurately. FEEINnet consists of three sub-networks: Feature Enhanced Mechanism (FEM) learns and extracts the fine-grained enhanced features from informative channels, the Edge Attention Mechanism (EAM) helps to precisely identify the edges of the lesion region and the INverse Attention Mechanism (INAM) generates inverse attention maps which emphasize the less confident or ambiguous regions thereby increasing the segmentation accuracy iteratively. These three sub-networks collectively help to improve feature extraction, enhance boundary detection, and refine segmentation maps, even in challenging scenarios with varying lesion sizes, shapes and pigmentation. FEEINnet consistently outperforms existing models, achieving a F1-score of 95.55%, 95.53%, and 94.52%; Intersection over Union (IoU) of 92.76%, 92.43%, and 91.34%; and Structural Similarity Index Measure (SSIM) of 94.63%, 93.51%, and 91.85% on the Human Against Machine 10000 (HAM10000), Pedro Hispano Hospital ( P H 2 ), and International Skin Imaging Collaboration 2018 (ISIC2018) datasets, respectively. The obtained results demonstrate that the proposed model has a greater ability to segment complex skin lesions more accurately.}, archive = {J_EAAI}, author = {Shivamm Warambhey and Aravindkumar Sekar}, doi = {10.1016/j.engappai.2025.112306}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112306}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-enhanced edge-INverse attention network for skin lesion segmentation}, volume = {162}, year = {2025}, }
• (2025). A severity indicator for unsupervised fault severity shift detection of heating, ventilation, and air conditioning sub-system faults using a novel adaptive feature weighing method. EAAI, 162, 112305. (www)

  @article{ , abstract = {Fault-free functioning of Heating, Ventilation, and Air Conditioning (HVAC) systems is essential for reducing energy waste in modern-day buildings. Hence, data-driven approaches for HVAC fault detection have gained popularity. Faults become more severe with time. Fault detection reveals the presence of an anomaly, but it does not convey how critical the fault severity is. Fault severity indication provides this essential context, enabling urgent resource allocation to more severe faults, adding practical significance. However, faults being rare, obtaining substantial data at different severity levels to train supervised Machine Learning models is a realistic challenge. Therefore, we propose a method for estimating fault severity in an unsupervised setting. We define a robust Severity Indicator (SI) that reflects the shift in the severity levels of a fault. First, we define a healthy domain boundary for fault-free data using One-Class Support Vector Machines. SI scores are then computed using a novel adaptive feature weighing algorithm that assigns weights to individual features, adaptively, for every fault. We focus on detecting the shift in severity, rather than quantifying it. The study of the robustness of SI for different faults in HVAC subsystems, chillers, and air handling units (AHUs) yields consistently promising results. Our comparative analysis shows that our method outperforms the unweighted approach and existing state-of-the-art techniques for fault severity estimation. Notably, our method excels in detecting low-severity faults, addressing a common limitation in current methods.}, archive = {J_EAAI}, author = {Ramnath V. Prabhu Bam and Rajesh S. Prabhu Gaonkar and Clint Pazhayidam George}, doi = {10.1016/j.engappai.2025.112305}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112305}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A severity indicator for unsupervised fault severity shift detection of heating, ventilation, and air conditioning sub-system faults using a novel adaptive feature weighing method}, volume = {162}, year = {2025}, }
• (2025). An efficient anchor-free model for ore particle size detection. EAAI, 162, 112304. (www)

  @article{ , abstract = {Accurate detection of ore size is crucial in mineral processing, directly impacting equipment efficiency and product quality. However, traditional anchor-based models often struggle with the irregular shapes and varying scales of ore particles, resulting in limited performance. To overcome these challenges, an anchor-free detection framework was proposed. It incorporates a cross-stage partial bottleneck and a spatial pyramid pooling cross-stage partial connections (SPPCSCP-DualConv), both enhanced with dual convolution, to improve feature extraction and multi-scale fusion. In the backbone, the dual convolution module combines group convolution with heterogeneous convolution to improve feature diversity. The SPPCSCP-DualConv module further enhances feature representation in complex backgrounds. Additionally, a simplified path aggregation network (simPANet) feature fusion module is employed in the neck to refine the integration of multi-scale features. The proposed model was trained using a combination of binary cross-entropy, complete intersection over union (IoU), and distribution focal loss to optimize detection accuracy. The proposed model achieved a mean average precision of 86.80 % at an IoU threshold of .5 and 78.50 % across IoU thresholds from .5 to .95, surpassing existing methods while maintaining a lightweight architecture with only 10.10 million parameters and 89.45 giga floating point operations per second. Ablation studies confirmed the effectiveness of the simPANet and SPPCSPC-DualConv modules in enhancing feature representation. Generalization tests across mining sites with similar distributions demonstrated strong performance, although limitations remain for exceptionally large ore blocks due to dataset bias. The proposed model significantly improved the accuracy and efficiency of ore particle size detection, providing reliable real-time insights to improve grinding control and mineral processing operations.}, archive = {J_EAAI}, author = {Kanghui Zhang and Qingkai Wang and Guobin Zou and Jiawei Yang and Tao Song and Yang Liu and Daoxi Liu}, doi = {10.1016/j.engappai.2025.112304}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112304}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient anchor-free model for ore particle size detection}, volume = {162}, year = {2025}, }
• (2025). Propeller fault-detection method for electric-propulsion aircraft using motor signals and generative model-based semi-supervised learning. EAAI, 162, 112301. (www)

  @article{ , abstract = {Failures in propulsion components, such as propellers, can critically affect flight safety; thus, early failure detection, preferably before flight, is essential. Traditional fault-diagnosis methods typically rely on additional sensors or operational data, which may not be available or practical in all situations. This study addresses these challenges by introducing motor-electric-signal-based fault diagnosis that is independent of airframe configuration and can detect faults, even when the aircraft is not in operation. However, difficulties arise owing to poor class variance in motor-electric-signal data and the challenge of obtaining fault data. To overcome these issues, a semi-supervised learning model based on a modified variational autoencoder-generative adversarial network (VAE-GAN) is proposed, which predicts faults using only normal motor-electric-signal data. Additionally, a new preprocessing method and patch-based ensemble inference technique are introduced to improve the poor class-variance characteristics of the data, thereby enhancing the prediction performance. This work demonstrates that propeller faults can be successfully diagnosed using motor-electric signals without the need for additional sensors or fault-data acquisition.}, archive = {J_EAAI}, author = {Sanga Lee and Dohyeong Kim and Minkyun Noh and Shinkyu Jeong and Jikang Kong and Youngjun Yoo}, doi = {10.1016/j.engappai.2025.112301}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112301}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Propeller fault-detection method for electric-propulsion aircraft using motor signals and generative model-based semi-supervised learning}, volume = {162}, year = {2025}, }
• (2025). A lightweight wood defect segmentation network via multi-dimension boundary perception and guidance. EAAI, 162, 112300. (www)

  @article{ , abstract = {Wood surface defect segmentation is extremely critical for defect refinement and quality control of wooden products. However, it is a challenging task to develop an efficient method with current algorithms due to the complicated characteristics of wood defects with obscure boundary, intraclass difference and interclass similarity. To address these issues, a lightweight network via multi-dimension boundary perception and guidance is proposed for precise segmentation of wood defects. At first, based on the Segformer, a boundary prediction branch is added to enrich detailed boundary information in the encoder, and supervised by the Gaussian signal and cosine similarity, to balance the effect of the boundary gradient information. Then, a double-flow enhancing module is designed to integrate the adjacent level features, by embedding two enhancing paths, to adaptively generate discriminative information of the defects. Finally, a binary segmentation head following the predicted map is introduced to strengthen the penalty for the false prediction results of the boundary. Experimental results demonstrate the proposed method outperforms the state-of-the-arts on our wood surface defect dataset, as well as on three public datasets.}, archive = {J_EAAI}, author = {Yuhang Zhu and Ye Lin and Zhezhuang Xu and Dan Chen and Kunxin Zheng and Yazhou Yuan}, doi = {10.1016/j.engappai.2025.112300}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112300}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight wood defect segmentation network via multi-dimension boundary perception and guidance}, volume = {162}, year = {2025}, }
• (2025). Sparse view tomographic reconstruction of elongated objects using learned primal-dual networks. EAAI, 162, 112295. (www)

  @article{ , abstract = {In the wood industry, logs are commonly quality screened by discrete X-ray scans on a moving conveyor belt from a few source positions. Typically, the measurements are obtained in a single two-dimensional (2D) plane (a “slice”) by a sequential scanning geometry. The data from each slice alone does not carry sufficient information for a three-dimensional tomographic reconstruction in which biological features of interest in the log are well preserved. In the present work, we propose a learned iterative reconstruction method based on the Learned Primal-Dual neural network, suited for sequential scanning geometries. Our method accumulates information between neighbouring slices, instead of only accounting for single slices during reconstruction. Evaluations were performed by training U-Nets on segmentation of knots (branches), which are crucial features in wood processing. Our quantitative and qualitative evaluations show that with as few as five source positions our method yields reconstructions of logs that are sufficiently accurate to identify biological features like knots (branches), heartwood and sapwood.}, archive = {J_EAAI}, author = {Buda Bajić and Johannes A.J. Huber and Benedikt Neyses and Linus Olofsson and Ozan Öktem}, doi = {10.1016/j.engappai.2025.112295}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112295}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sparse view tomographic reconstruction of elongated objects using learned primal-dual networks}, volume = {162}, year = {2025}, }
• (2025). A deep Q-network-driven multi-objective evolutionary algorithm for distributed heterogeneous hybrid flow shop scheduling with worker fatigue. EAAI, 162, 112292. (www)

  @article{ , abstract = {The hybrid flow shop scheduling problem (HFSP) is a prominent challenge in advanced manufacturing systems. Existing research often overlooks the impact of workers in production shops or treats worker fatigue as a static parameter, failing to capture its nonlinear accumulation and recovery effects on processing efficiency. However, with the advent of Industry 5.0, there has been a growing emphasis on the critical role of human factors in production scheduling. As a result, designing an effective algorithm for HFSP that considers human factors has become a prominent research focus. In this paper, an extended distributed heterogeneous hybrid flow shop scheduling problem with the dynamic effects of worker fatigue (DHHFSP-WF) is investigated. To address this problem, a Deep Q-Network-based multi-objective optimization algorithm (DQNMOEA) is designed to minimize makespan, total energy consumption (TEC), and total worker idle time (WIT). In DQNMOEA, a four-dimensional vector encoding scheme considering worker allocation represents solution, and a reconstruction strategy ensures initial population quality and diversity. Moreover, an improved order crossover, two-point crossover, and a segment-based recombination mutation method are proposed to enhance the global search performance of the algorithm. Then, a problem-specific local search strategy is designed for each layer of the vector, allowing the Deep Q-Network (DQN)-based adaptive decision-making mechanism to perform local perturbations on the current non-dominated solutions in the most suitable dimensions. Finally, seven algorithms are adopted to make a comparison on 36 sets of instances, the experimental results indicate that DQNMOEA exhibits competitive performance in solving DHHFSP-WF.}, archive = {J_EAAI}, author = {Jianlin Zhang and Longbin Ma and Wu Zhao and Jie Cao and Zuohan Chen and Tianpeng Xu}, doi = {10.1016/j.engappai.2025.112292}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112292}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep Q-network-driven multi-objective evolutionary algorithm for distributed heterogeneous hybrid flow shop scheduling with worker fatigue}, volume = {162}, year = {2025}, }
• (2025). An angle-enhanced deep learning framework for thermal defect diagnosis in overhead transmission line composite insulators. EAAI, 162, 112285. (www)

  @article{ , abstract = {Due to long-term outdoor exposure, composite insulators are susceptible to degradation and abnormal temperature rise, making Unmanned Aerial Vehicle (UAV)-based infrared inspections essential for effective monitoring. However, traditional manual interpretation of these images is inefficient and subjective. To improve detection automation and accuracy, we propose an intelligent detection method for composite insulators in infrared images based on an improved You Only Look Once version 11 (YOLOv11) model. The proposed approach introduces Oriented Bounding Boxes (OBBs) for annotation and designs an Angle-Enhanced Probabilistic Intersection over Union (AE-ProbIoU) loss function to enhance the model's ability to detect rotated objects. Experimental results demonstrate that the proposed Angle-Enhanced You Only Look Once (AE-YOLO) model achieves a mAP50:95 of 94.0 % and an angle prediction accuracy of 94.3 %. In addition, a temperature extraction module based on the OBBs is developed to accurately derive the temperature profile of the insulator core rod. This method significantly enhances the intelligence level of infrared image analysis for composite insulators and provides technical support for condition assessment and fault prediction in power transmission lines.}, archive = {J_EAAI}, author = {Xinzhe Yu and Zhenan Zhou and Yu Deng and Kun Zhang and Chen Gu and Zheyuan Liu and Songsong Zhou}, doi = {10.1016/j.engappai.2025.112285}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112285}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An angle-enhanced deep learning framework for thermal defect diagnosis in overhead transmission line composite insulators}, volume = {162}, year = {2025}, }
• (2025). Bayesian bidirectional long short-term memory-based kinematics-dynamics fusion for fault-tolerant vehicle state estimation under yaw rate sensor failures. EAAI, 162, 112283. (www)

  @article{ , abstract = {Accurate estimation of vehicle states is a fundamental component of vehicle stability control systems. To address the issue of inaccurate estimation of vehicle state parameters resulting from yaw rate sensor failures, this study proposes a three-mode collaborative fault-tolerant state estimation method based on Bayesian Bidirectional Long Short-Term Memory (BiLSTM) kinematics-dynamics fusion. First, the kinematics-based method is established using the kinematics model. Second, the dynamics-based method is designed by integrating the Unscented Kalman Filter (UKF) with the dynamics model. Subsequently, a BiLSTM network fusion model based on Bayesian optimization is presented. The model utilizes estimates from kinematic and kinetic methods as a priori inputs and combines the bidirectional information capturing capability of BiLSTM with hyperparameter tuning from Bayesian optimization. The results indicate that when the yaw rate sensor fails, the proposed method achieves an average Root Mean Square Error (RMSE) of 0.0276 km per hour (km/h) for longitudinal speed, 0.0008 radian (rad) for side slip angle, and 0.0072 radian per second (rad/s) for yaw rate across all scenarios. This performance demonstrates a superiority over various maneuvers. This paper combines kinematics, dynamics, and deep learning to provide a reliable solution for fault-tolerant estimation of vehicle states.}, archive = {J_EAAI}, author = {Min Gao and Jiaqi Li and Wei Wang and Renguang Wang and Jin Luo and Jing Li}, doi = {10.1016/j.engappai.2025.112283}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112283}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bayesian bidirectional long short-term memory-based kinematics-dynamics fusion for fault-tolerant vehicle state estimation under yaw rate sensor failures}, volume = {162}, year = {2025}, }
• (2025). Hierarchical binary diffing with inductive code representation learning with graph sampling-and-aggregate. EAAI, 162, 112279. (www)

  @article{ , abstract = {Binary program diffing, or simply binary diffing, is a type of program analysis technique that quantifies the similarity between two binary programs to derive their differences. In particular, binary diffing is an essential technique for uncovering vulnerabilities and potential attack vectors in industrial control systems, where patch deployment is complicated by closed and restricted environments. Studies on binary diffing can be broadly categorized into dynamic analysis-based, static analysis-based, and neural network-based approaches. Each category of existing studies has its shortcomings, including limited coverage, low accuracy, and issues with on-demand learning. In this paper, we propose the binary diffing with sampling-and-aggregate, a hierarchical binary diffing model that generates inductive code representations based on graph sampling-and-aggregate. Our model sequentially produces instruction-level embedding, block-level embedding, and function-level embedding from the inter-procedural control flow graph of a given program, and then performs hierarchical code diffing based on these embeddings. We formally define the detailed models and present the algorithm of hierarchical binary diffing. Additionally, we conduct a thorough analysis of this algorithm, deriving several advantages. We implemented a prototype and evaluated it on a large-scale dataset in a cross-version, cross-optimization, and obfuscation settings. Our prototype showed F1-scores up to 0.96 and 0.968 in cross-version setting for function and basic block diffing, respectively. Also, our method demonstrated its robustness over several binary obfuscations. In conclusion, our proposal, which generates basic block- and function-level embedding by considering the control flow, has solid advantages on binary diffing and shows the robustness on the binary tampering.}, archive = {J_EAAI}, author = {Seungho Jeon and Kijong Koo and Daesung Moon and Jung Taek Seo}, doi = {10.1016/j.engappai.2025.112279}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112279}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical binary diffing with inductive code representation learning with graph sampling-and-aggregate}, volume = {162}, year = {2025}, }
• (2025). Nonlinear excavation load prediction of hydraulic excavator based on gated recurrent unit neural network. EAAI, 162, 112278. (www)

  @article{ , abstract = {Excavator arms are integral to the mining and construction industries, where real-time excavation load prediction is a critical element for the advancement of automated excavation technology. This study presents a novel Physics-guided Neural Network (PGNN) designed to predict the excavating force of hydraulic cylinders used in earthwork excavation. The PGNN model synergizes the physical load model of excavators with a Gated Recurrent Unit (GRU) neural network and is optimized using the Hyperband algorithm to attain both high-speed and precise forecasting. Through comparative experiments, the study validates the PGNN model's ability to achieve optimal response speed and precision in predicting excavation loads. Additionally, the predictive performance of the PGNN model is assessed via a Hardware-in-the-loop (HIL) test, conducted within the context of an actual excavation experiment. This research introduces a promising approach that seamlessly integrates physics-based modeling with machine learning techniques, facilitating real-time load forecasting for excavators. The findings pave the way for more efficient and precise excavation processes, with implications for the broader fields of mining and construction automation.}, archive = {J_EAAI}, author = {Jinshi Chen and Yue Yu and Dongyang Huo and Han Zhang and Jingyan Wang}, doi = {10.1016/j.engappai.2025.112278}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112278}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Nonlinear excavation load prediction of hydraulic excavator based on gated recurrent unit neural network}, volume = {162}, year = {2025}, }
• (2025). Plug-and-play fine grained neural cognitive diagnosis framework. EAAI, 162, 112276. (www)

  @article{ , abstract = {Cognitive diagnosis (CD) is a core task in intelligent education, which accurately assesses students’ mastery of specific knowledge concepts (KCs) by analyzing their answer records. However, existing methods mainly rely on explicit interaction data and use diagnostic models for automatic knowledge proficiency inference. These methods lack systematic optimization for fine-grained knowledge level representation, making it difficult to fully reflect students’ true learning status. To address this, this paper introduces a Plug-and-Play F ine Grained N eural C ognitive D iagnosis Framework (FNCD) with Knowledge-Level Constraint Awareness . The framework combines a knowledge proficiency evaluation module with students’ answer records and a Q-matrix to statically assess knowledge mastery. It uses a student similarity construction method based on random grouping to reveal latent learning pattern associations. Additionally, it employs a multi-scale relational learning strategy and a Top-k attention-enhanced graph network mechanism to dynamically adjust the student similarity relationship network, accurately modeling the complex learning relationships between students. Ultimately, a joint training mechanism is used to optimize the outputs of each module, significantly improving the rationality, interpretability, and accuracy of CD. The experimental results demonstrate that FNCD, as an artificial intelligence-driven plug-and-play module, can be effectively integrated into existing CD models to enhance the modeling of fine-grained knowledge mastery and improve diagnostic accuracy, showcasing the application potential of artificial intelligence in personalized education.}, archive = {J_EAAI}, author = {Xinjie Sun and Qi Liu and Kai Zhang and Weiyin Gong and Shuanghong Shen and Fei Wang and Yan Zhuang and Meikai Bao and Shijin Wang and Yuling Ma and Enhong Chen}, doi = {10.1016/j.engappai.2025.112276}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112276}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Plug-and-play fine grained neural cognitive diagnosis framework}, volume = {162}, year = {2025}, }
• (2025). Open-vocabulary object detection via neighboring region attention alignment. EAAI, 162, 112270. (www)

  @article{ , abstract = {The nature of diversity in real-world environments necessitates neural network models to expand from closed category settings to accommodate novel emerging categories. In this paper, we study open-vocabulary object detection (OVD), which facilitates the detection of novel object classes under the supervision of only base annotations and open-vocabulary knowledge. However, we find that the inadequacy of distilled information from the detector head during the alignment process inevitably constrains the performance of recent distillation-based OVD strategies. To this end, we propose Neighboring Region Attention Alignment (NRAA), which performs alignment within the attention mechanism to boost the open-vocabulary inference. Specifically, for a given proposal region, we randomly explore the neighboring boxes to capture the surrounding contextual vocabulary knowledge. Then, a set of regional token features, encompassing both the proposal and neighboring regions, utilize our proposed Neighboring Region Attention (NRA) to extract interaction information. Finally, this information is seamlessly provided to the distillation procedure to assist the alignment between the detector and the pre-trained vision-language models (VLMs). Extensive experiments validate that our proposed model exhibits superior performance on open-vocabulary benchmarks.}, archive = {J_EAAI}, author = {Sunyuan Qiang and Xianfei Li and Yanyan Liang and Wenlong Liao and Tao He and Pai Peng}, doi = {10.1016/j.engappai.2025.112270}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112270}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Open-vocabulary object detection via neighboring region attention alignment}, volume = {162}, year = {2025}, }
• (2025). A time and frequency convolutional autoencoder for anomaly detection in industrial robots based on inertial measurement unit error calibration. EAAI, 162, 112269. (www)

  @article{ , abstract = {In the realm of industrial robots, ensuring operational reliability and Long-Term Autonomy hinges on the accurate detection of anomalies. However, this sample difference due to noise, joint random errors and sensor errors increases the challenge of robot anomaly detection. To address this problem, an unsupervised deep learning method based on inertial measurement unit (IMU) error calibration is proposed. Firstly, the attitude signals acquired by the IMU from the end of the robot were calibrated using Kalman filtering. The three dimensional (3D) free acceleration was corrected based on the calibrated attitude signal and the calibrated 3D free acceleration signal was used as a signal sample. Secondly, a time and frequency convolutional autoencoder model (TFCAE) is proposed. And the distribution of the different component signals is fitted by stacking multiple encoder modules and 3D-TFCAE is used for 3D free acceleration signal reconstruction model. Then, the error sphere radius is calculated based on the reconstruction error of the 3D free acceleration signal. And the error sphere radius is used as the anomaly detection threshold to realize the robust detection of different types of anomalies. The model was evaluated on a constructed anomaly dataset. This study contributes an innovative 3D-TFCAE architecture, integrating Kalman filtering with time-frequency feature fusion, markedly enhancing anomaly detection in complex signal environments. Experimental findings reveal that 3D-TFCAE significantly outperforms 18 baseline models, improving detection accuracy by about 20 %–40 %, offering an effective solution for high-precision anomaly detection in industrial robots. The code for this project is available at https://github.com/LJlong977/3DTFCAE .}, archive = {J_EAAI}, author = {Jianlong Li and Xiaoqin Liu and Xing Wu and Dongxiao Wang and Kai Xu and Yashan Li}, doi = {10.1016/j.engappai.2025.112269}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112269}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A time and frequency convolutional autoencoder for anomaly detection in industrial robots based on inertial measurement unit error calibration}, volume = {162}, year = {2025}, }
• (2025). A lightweight citrus detection and counting method based on deep learning model. EAAI, 162, 112268. (www)

  @article{ , abstract = {Picking robots have become an important development direction of smart agriculture, and the accurate detection, counting and lightweight deployment of fruits are the technical basis for realizing robot picking. However, due to complex weather conditions and the possible mutual occlusion between branches and citrus, it is challenging to accurately detect and count citrus in orchards. This study proposes a lightweight small target detection model for detecting and counting citrus, and deploys it on the citrus detection platform. The model first introduces FasterNet Block into the cross-stage partial feature fusion module of the backbone network to reduce the number of parameters and calculations while improving the detection accuracy of the network. Secondly, a multi-scale attention mechanism is added to the backbone network to enhance the feature extraction ability of the network. Finally, a bounding box loss function based on a dynamic non-monotonic focusing mechanism is used to increase the model convergence speed and further improve the model accuracy. Experimental results show that the model has an accuracy of 92.7%, an average precision of 91.7%, and a model size of only 5.37 megabytes. The lightweight model is applied to the citrus detection platform. Based on this application, a citrus counting method is proposed, which obtains a mean absolute error (MAE) of 0.92, a root mean square error (RMSE) of 1.28, a determination coefficient ( R 2 ) of 0.98, and a frame rate of 80.6 per second, which meets the requirements of real-time citrus detection and counting. This provides technical support for the subsequent deployment and counting research of picking robots.}, archive = {J_EAAI}, author = {Jiqing Chen and Mingchang Zhang and Bin Lu and Quan Chen and Zhiwu Jiang and Peilin Li and Jingyao Gai}, doi = {10.1016/j.engappai.2025.112268}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112268}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight citrus detection and counting method based on deep learning model}, volume = {162}, year = {2025}, }
• (2025). Medical image fusion for high-level analysis: A mutual enhancement framework for unaligned photoacoustic tomography and magnetic resonance imaging. EAAI, 162, 112259. (www)

  @article{ , abstract = {Photoacoustic tomography (PAT) offers optical contrast, whereas magnetic resonance imaging (MRI) excels in imaging soft tissue and organ anatomy. The fusion of PAT with MRI holds promising application prospects due to their complementary advantages. Existing image fusion has made considerable progress in pre-registered images, yet spatial deformations are difficult to avoid in medical imaging scenarios. More importantly, current algorithms focus on visual quality and statistical metrics, thus overlooking the requirements of high-level tasks. To address these challenges, we propose an unsupervised fusion model, termed PAMRFuse+, which integrates image generation and registration. Specifically, a cross-modal style transfer network is introduced to simplify cross-modal registration to single-modal registration. Subsequently, a multi-level registration network is employed to predict displacement vector fields. Furthermore, a dual-branch feature decomposition fusion network is proposed to address the challenges of cross-modal feature modeling and decomposition by integrating modality-specific and modality-shared features. PAMRFuse+ achieves satisfactory results in registering and fusing unaligned PAT–MRI datasets. Moreover, for the first time, we evaluate the performance of medical image fusion with multi-organ instance segmentation. Extensive experimental demonstrations reveal the advantages of PAMRFuse+ in improving the performance of medical image analysis tasks.}, archive = {J_EAAI}, author = {Yutian Zhong and Jinchuan He and Zhichao Liang and Shuangyang Zhang and Qianjin Feng and Lijun Lu and Li Qi}, doi = {10.1016/j.engappai.2025.112259}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112259}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Medical image fusion for high-level analysis: A mutual enhancement framework for unaligned photoacoustic tomography and magnetic resonance imaging}, volume = {162}, year = {2025}, }
• (2025). Stacked machine learning-based probabilistic seismic demand and risk assessment of railway embankments with variable slopes. EAAI, 162, 112245. (www)

  @article{ , abstract = {Accurate seismic risk assessment of railway embankments is critical for risk mitigation, seismic design, and emergency planning. However, conventional methods often suffer from computational inefficiency and complexity. This study proposes a novel machine learning (ML) framework to rapidly and accurately evaluate probabilistic seismic demand and risk for railway embankments. Latin hypercube sampling is utilised to generate representative soil parameter samples to construct numerical models for simulating dynamic responses under near-fault pulse-like ground motions. The peak permanent settlement (PPS) of the embankment surface is used as the key performance metric. Multiple ML models, including decision trees, random forests (RFs), extreme gradient boosting (XGBoost), artificial neural networks (ANNs), and a stacked ML model that integrates RFs, XGBoost, and ANNs, are trained and compared. The stacked ML model outperforms the other models and achieves the highest predictive accuracy for the PPS. SHapley Additive exPlanations are used to identify the velocity spectrum intensity (VSI) and the internal friction angle of the embankment as the most influential factors. Seismic fragility and risk curves are subsequently developed. The VSI and a power-law seismic hazard function are combined to estimate the annual exceedance probabilities for three seismic design criteria levels. The proposed ML framework significantly enhances the efficiency of seismic risk analysis while maintaining high precision, thereby providing a transformative approach for the seismic assessment of railway embankments.}, archive = {J_EAAI}, author = {Pan Si and Liang Tang and Shuang Tian and Xianzhang Ling and Yanfang Liu}, doi = {10.1016/j.engappai.2025.112245}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112245}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stacked machine learning-based probabilistic seismic demand and risk assessment of railway embankments with variable slopes}, volume = {162}, year = {2025}, }
• (2025). Adaptive chemical industrial processes fault detection model based on sparse filtering-based improved mixed-gaussian probabilistic principal component analysis considering low-probability events. EAAI, 162, 112236. (www)

  @article{ , abstract = {Probabilistic Principal Component Analysis (PPCA) is widely used in process monitoring. However, its underlying assumption that data follows a Gaussian distribution limits its effectiveness in handling Low Probability Events (LPEs), which often deviate from this assumption. To address this challenge, we propose a novel method called Sparse Filtering-based Improved Mixed-Gaussian Probabilistic Principal Component Analysis (SFIMPPCA) for enhanced LPEs detection. First, a Sparse Filtering (SF) preprocessing technique with an incremental structure is employed to extract the most discriminative features. Second, to address the distortion caused by LPEs, a dynamic ratio correction mechanism based on statistical variability is introduced, followed by a newly designed Mixed-Gaussian Probabilistic Principal Component Analysis (MPPCA). Third, a Bayesian Optimization Algorithm (BOA) is applied to automatically adjust control limits, enhancing the accuracy and reliability of fault detection. The effectiveness of the proposed method is validated using the Tennessee Eastman (TE) process and the Tin Chemical Process (TCP). Experimental results demonstrate that the proposed method significantly improves performance under LPEs conditions, achieving a 10%–12% improvement in most cases.}, archive = {J_EAAI}, author = {Chuangyan Yang and Jiande Wu and Peng Li and Xun Lang and Mingxi Ai and Hancheng Wang}, doi = {10.1016/j.engappai.2025.112236}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112236}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive chemical industrial processes fault detection model based on sparse filtering-based improved mixed-gaussian probabilistic principal component analysis considering low-probability events}, volume = {162}, year = {2025}, }
• (2025). Dynamic graph neural network with historic-sequential information representation. EAAI, 162, 112234. (www)

  @article{ , abstract = {In today’s interconnected world, networks, whether social, technological, or biological, are constantly evolving, with relationships forming, shifting, and dissolving over time. Traditional models struggle to capture this fluidity, treating networks as static snapshots rather than living systems. While existing research has made strides in analyzing either spatial–temporal patterns or temporal sequences separately, these approaches often fall short in scalability and fail to unify discrete and continuous-time perspectives. To bridge this gap, we introduce DyGHS (Dynamic Graph Historical-Sequential), an innovative framework that harnesses dynamic graph neural networks to model how nodes, edges, and their historical interactions evolve together. By seamlessly integrating discrete-time and continuous-time graph representations, DyGHS not only captures the richness of real-world networks but also efficiently predicts future connections and node behaviors. Our experiments reveal that combining continuous-time Fourier transform (CTFT) with graph neural networks significantly boosts prediction accuracy, outperforming current methods in tasks like link prediction and node classification. This advancement opens new doors for understanding and anticipating the ever-changing tapestry of networked systems.}, archive = {J_EAAI}, author = {Adam Abakar Hamid and Anping Zhao and Alladoumbaye Ngueilbaye}, doi = {10.1016/j.engappai.2025.112234}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112234}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic graph neural network with historic-sequential information representation}, volume = {162}, year = {2025}, }
• (2025). MAE-diff: Masked-AutoEncoder-guided diffusion framework for source-free domain adaptive medical image segmentation. EAAI, 162, 112219. (www)

  @article{ , abstract = {Domain gaps can often cause dramatic performance deterioration when applying medical image segmentation models trained on the source domain to the target domain. Although unsupervised domain adaptation methods can address the domain gap challenge to some extent, their reliance on accessing source images largely hampers their practical applicability, as source data are often inaccessible due to privacy concerns. Moreover, the low-quality characteristic of medical images can further degrade the domain adaptation performance of segmentation models. To address these issues, we propose the Masked-AutoEncoder-guided Diffusion (MAE-Diff) framework for source-free domain adaptive medical image segmentation. MAE-Diff mainly consists of a Masked AutoEncoder (MAE) Module for effective feature extraction and domain adaptation, and a Diffusion Module for effective segmentation of low-quality medical images. On source images, the MAE encoder is trained to extract image-specific features, and the Diffusion Module is trained to generate segmentation maps following a gradual denoising strategy, under the guidance of features extracted by the MAE encoder. Training on the target domain involves only fine-tuning MAE (trained on the source images) with target images, allowing MAE-Diff to adapt to the target domain distribution. Inference on target images can then be made by the source-based Diffusion Module, under the guidance of features extracted by the MAE encoder fine-tuned on the target images. Extensive experiments on three datasets demonstrate the effectiveness of the proposed framework for source-free domain-adaptive medical image segmentation. The code of MAE-Diff is available at https://github.com/xuss804/MAEDiff .}, archive = {J_EAAI}, author = {Shanshan Xu and Le Xu and Yeqing Yang and Lixia Tian}, doi = {10.1016/j.engappai.2025.112219}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112219}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MAE-diff: Masked-AutoEncoder-guided diffusion framework for source-free domain adaptive medical image segmentation}, volume = {162}, year = {2025}, }
• (2025). Dynamic-static siamese Takagi–Sugeno-kang fuzzy system with inductive-reflection deep fuzzy rule. EAAI, 162, 112199. (www)

  @article{ , abstract = {The first-order Takagi–Sugeno–Kang (TSK) fuzzy classifiers are famous for their high computational efficiency and strong interpretability, but they often struggle to learn from complex and large-scale datasets, and perform not very well compared to higher-order TSK fuzzy classifiers. To address this issue, in this paper we propose a novel Dynamic-Static Siamese TSK Fuzzy classifier with Inductive-Reflection Deep Fuzzy rule. It aims to enhance the model’s self-learning capabilities by utilizing Siamese network to integrate deep fuzzy knowledge and fine-grained knowledge without the need for a teacher model. The innovations of this study are as follows: (1) The deep fuzzy rules in the proposed classifier are enriched with an “Inductive-Reflection” process, which reduces constraints on traditional basic fuzzy rule and aligns rule acquisition more closely with general human thinking manners; (2) The proposed method includes a mechanism for self-learning and improvement from both deep and fine-grained fuzzy knowledge, eliminating the complexity of retraining a new teacher model; (3) An adaptive learning function is developed to effectively adjust the learning process, adapting to tasks with different complexities. Extensive experiments results on benchmark datasets, as well as two real-world datasets, demonstrate the effectiveness of the proposed classifier in terms of classification accuracy and weighted F1-score.}, archive = {J_EAAI}, author = {Xiongtao Zhang and Qihuan Shi and Yunliang Jiang and Qing Shen and Jungang Lou and Ruiqin Wang}, doi = {10.1016/j.engappai.2025.112199}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112199}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic-static siamese Takagi–Sugeno-kang fuzzy system with inductive-reflection deep fuzzy rule}, volume = {162}, year = {2025}, }
• (2025). Enhancing anomaly detection with few-shot fine-tuned long text-to-image models. EAAI, 162, 112174. (www)

  @article{ , abstract = {Industrial anomaly detection plays a crucial role in the industrial manufacturing field. Currently, utilizing generated data to improve the performance of the anomaly detection model is an effective approach. However, most existing methods often rely on mask-guided synthesis, where the distribution of the generated defects is limited by masks that are typically random or learned by a model. In addition, the scarcity of real anomalous samples makes it difficult for generative models to capture genuine defect patterns and align with the real anomaly distribution. To tackle these issues, we propose DefectGen, the first long-text-guided few-shot text-to-image data generation pipeline for industrial anomaly detection. To improve distribution alignment under limited anomaly samples, DefectGen incorporates a Prompt Generation and Variation Module, which uses MLLMs (Multimodal Large Language Models) to expand few-shot image–text pairs into diverse and semantically rich prompts, and DoKr (Weight- D ecomposed L o w-Rank Adaptation with Kr onecker product), a lightweight fine-tuning strategy with structured low-rank adaptation. To ensure the quality of synthetic data, DefectGen further introduces the Real-Guided Clustering Filter, which selects high-quality generated samples by comparing their features with those of real anomalies. Experiments on the MVTec AD(MVTec AnomalyDetection) dataset show that DefectGen generates more diverse and realistic synthetic anomalies and achieves a 5.58% average improvement in anomaly classification accuracy compared to state-of-the-art methods. Code and data are available at: https://anonymous.4open.science/r/DefectGen-CD04/ .}, archive = {J_EAAI}, author = {Jiachen Liu and Jiajia An and Junbin Lu and Zhuoqin Yang and Jinbao Wang and Ping Lu and Yuying Wang and Linlin Shen}, doi = {10.1016/j.engappai.2025.112174}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112174}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing anomaly detection with few-shot fine-tuned long text-to-image models}, volume = {162}, year = {2025}, }
• (2025). Physics-informed fine-tuning for physics discovery from random and sparse data. EAAI, 162, 112132. (www)

  @article{ , abstract = {Although significant advances have been made in both numerical methods and machine learning approaches for solving differential equations across various scientific sectors, these methods often rely on complete information about the differential equations, including precise parameter values, which are not easily obtainable in real-world scenarios. To address this challenge, data-driven methods for discovering differential equations have gained growing popularity in recent years. However, many existing approaches demand unrealistic prerequisites, such as extensive high-fidelity data or carefully designed low-fidelity data and functions. In this paper, we propose a novel method: P hysics- I nformed F ine- T uning ( PIFT ) to discover the unknown parameters in differential equations when only randomly distributed sparse data points are available. PIFT consists of three stages; (i) generating low fidelity data from prior knowledge under realistic settings, (ii) pre-training a single neural network with the generated low fidelity data, and (iii) fine-tuning the pre-trained model using physics-informed loss function. PIFT is evaluated on seven scientific problems including five ordinary differential equations and two partial differential equations. We also demonstrate the robustness and generalizability of PIFT to out-of-distribution tasks. PIFT exhibits high accuracy and robustness in discovering unknown parameters of differential equations from randomly distributed sparse data points.}, archive = {J_EAAI}, author = {Yong Jin Jeong and Taesup Moon}, doi = {10.1016/j.engappai.2025.112132}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112132}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed fine-tuning for physics discovery from random and sparse data}, volume = {162}, year = {2025}, }
• (2025). STEval: A framework for evaluating spatio-temporal crime prediction models. EAAI, 162, 112123. (www)

  @article{ , abstract = {Spatio-Temporal predictive models are crucial for forecasting where and when crimes will occur, aiding public security organizations in resource allocation and crime prevention. Despite numerous literature proposals, a lack of standardized evaluation criteria hinders comparability and reliability. To address this, we propose STEval, a comprehensive and flexible evaluation framework for spatio-temporal predictive models. STEval consists of four modules: data preparation, spatial structure definition, model training, and model evaluation. The framework’s robustness was demonstrated using a 40-million-record crime dataset from Minas Gerais State (Brazil) across five experimental scenarios, including variations in temporal granularity, spatial resolution, and distribution shifts. Results demonstrate that no single model is universally superior, and the most appropriate model depends on the specific application context. For instance, Spatio-Temporal Kernel Density Estimation (STKDE) consistently achieved high Hit Rates (HR), often exceeding 0.98 in fine spatial resolutions (e.g., 100-square meters grid) for violent crimes, while Spatial-Temporal Autoregressive Integrated Moving Average (STARIMA) demonstrated strong performance in temporal granularity tests, reaching HRs of up to 0.65 for theft predictions. Conversely, Extra Tree Regressor, though exhibiting significantly lower HRs (e.g., as low as 0.02 in some spatial tests), consistently provided the fastest execution times, often under 5 s, contrasting with STKDE’s execution times that could extend to thousands of seconds in dense spatial grids. The framework’s detailed analysis provides important insights for informed model selection and optimization, revealing each model’s strengths and limitations, and providing a robust foundation for future advancements in spatio-temporal crime prediction research, leveraging Machine Learning (ML) techniques.}, archive = {J_EAAI}, author = {Gabriel Amarante and Matheus Pimenta and Yan Andrade and Matheus Senna and Rainer Menezes and Antônio Hot Faria and Marcelo Vilas-Boas and Frederico Martins de Paula Neto and João Paulo da Silva and Everton Renato de Sousa and Jamicel da Silva and Wagner Meira Jr. and George Teodoro and Leonardo Rocha and Renato Ferreira}, doi = {10.1016/j.engappai.2025.112123}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112123}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {STEval: A framework for evaluating spatio-temporal crime prediction models}, volume = {162}, year = {2025}, }
• (2025). A new design of arithmetic and logic unit for enhancing the security of future internet of things devices using quantum-dot technology. EAAI, 162, 112113. (www)

  @article{ , abstract = {The Internet of Things (IoT) is a network of interconnected devices that collect, monitor, analyze, and exchange data. This technology plays a crucial role in the smart city infrastructure by seamlessly interconnecting various nodes. The extensive application and recognition of IoT across multiple city domains, such as healthcare, transportation, energy, education, and agriculture, bring significant challenges, with security among the most pressing. Traditional hardware technologies like Complementary Metal Oxide Semiconductor (CMOS) and Very Large Scale Integration (VLSI) suffer from limitations such as high power consumption and insufficient scalability, which hinder secure and sustainable IoT deployment. Such limitations have prompted the need to seek other technologies that would serve the dual purpose of providing security as well as energy. Quantum-based technologies can become adequate candidates offering promising solutions to make IoT devices and sustainable systems more secured. Quantum-dot Cellular Automata (QCA) has been proposed as a nanotechnology with the potential of consuming ultra-low powers, less area, and high-speed operation. QCA enhances security through sustainable computing objectives by minimizing energy usage. To improve the future security and efficiency of IoT hardware, this paper suggests a QCA-based Arithmetic Logic Unit (ALU). This ALU can generate more than 12 logical and arithmetic operations. Designed together with the majority gates, XOR gates, multiplexers, and full adders, the ALU is simulated using the QCA-Designer 2.0.3. Simulated results indicate improvements in the number of cells and reduced occupied area relative to the earlier designs. These results indicate the potential of QCA technology in enabling secure, energy-efficient, and compact computing architecture applicable in the future IoT.}, archive = {J_EAAI}, author = {Maryam Zaker and Seyed Sajad Ahmadpour and Nima Jafari Navimipour and Muhammad Zohaib and Neeraj Kumar Misra and Sankit Kassa and Ahmad Habibizad Navin and Arash Heidari and Mehdi Hosseinzadeh and Omar I. Alsaleh}, doi = {10.1016/j.engappai.2025.112113}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112113}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new design of arithmetic and logic unit for enhancing the security of future internet of things devices using quantum-dot technology}, volume = {162}, year = {2025}, }
• (2025). GhostPointNet: A deep learning-based method for ghost point noise detection in four-dimensional (4D) millimeter-wave radar point clouds of underground mine. EAAI, 161, 112380. (www)

  @article{ , abstract = {The high dust concentration, multi-metal supports, and narrow winding tunnels in underground mines collectively lead to frequent ghost point noise in four-dimensional (4D) millimeter-wave radar point clouds, posing serious challenges for mining perception and localization. To address this, we propose a deep learning algorithm, named GhostPointNet, for 4D millimeter-wave radar ghost point detection in underground mining environments. From an artificial intelligence perspective, this model thoroughly considers the multi-modal features of 4D millimeter-wave radar and the environmental complexity of underground mines. It incorporates multi-parameterized spatial information inputs in both Cartesian and Spherical coordinates, coupled with “Double T-Net” adaptive alignment correction, while integrating non-spatial information such as radar power and Doppler data to achieve multi-modal representation and end-to-end discrimination between ghost points and real points. Experimental validation shows that GhostPointNet achieves excellent performance in underground mining scenarios with 92.45 % accuracy and 95.84 % F1-score, outperforming traditional filtering, clustering, and machine learning algorithms. From an engineering application perspective, GhostPointNet is specifically designed for ghost noise detection in underground mines. Even in complex scenarios such as mine tunnel intersections and turns, it preserves critical structural points. Its end-to-end neural network simplifies post-processing procedures, enhances operational efficiency, and provides stable and reliable perceptual support for subsequent tasks such as autonomous mine locomotive navigation and three-dimensional (3D) structure reconstruction. Experimental results demonstrate that this method surpasses baseline approaches in ghost point detection, real point preservation, and generalization capability, providing significant support for improving underground mining safety and efficiency.}, archive = {J_EAAI}, author = {Hu Liu and Zhenghua Zhang and Jing Yang and Jörg Benndorf and Xiaofei Wang and Jiaqi Dong and Zitao Lin and Guoliang Chen}, doi = {10.1016/j.engappai.2025.112380}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112380}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GhostPointNet: A deep learning-based method for ghost point noise detection in four-dimensional (4D) millimeter-wave radar point clouds of underground mine}, volume = {161}, year = {2025}, }
• (2025). Hybrid ensemble learning for predicting peak deviatoric stress in soil-rock mixtures from triaxial test data. EAAI, 161, 112377. (www)

  @article{ , abstract = {Peak deviatoric stress ( q f ) is an important mechanical property index of soil-rock mixtures (SRM). However, determining it through experiments is a challenging and time-consuming task. This study presents a weighted averaging ensemble model for predicting the q f of SRM. 585 sets of consolidated drained triaxial test data for SRM were collected to construct the proposed models. Firstly, taking the extreme gradient boosting (XGBoost) as an example, the accuracy of the models established with different input parameter combinations was tested to determine the optimal inputs. Then, five artificial intelligence models were used to train the dataset, and the Bayesian optimization method was adopted for hyperparameter adjustment. Based on the prediction results, two base models (i.e. XGBoost and random forest (RF)) with good performance were selected from these five models, and a novel weighted averaging ensemble model was developed to predict the q f of SRM. The coefficient of determination (R 2 ), root mean squared error (RMSE) and mean absolute error (MAE) values of the ensemble model were 0.990, 220.0 kPa (kPa) and 118.4 kPa, respectively, for all datasets, indicating that this model has great potential in predicting the q f of SRM. In addition, the robustness, parameter patterns and the SHapley Additive exPlanations (SHAP) analysis of the ensemble model were also conducted in this study. The results of this study can be applied to rapid analysis of mechanical parameters for SRM in engineering, and are of great significance for geotechnical engineering design and disaster prevention.}, archive = {J_EAAI}, author = {Ruiliang Zhang and Xinhua Xue}, doi = {10.1016/j.engappai.2025.112377}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112377}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid ensemble learning for predicting peak deviatoric stress in soil-rock mixtures from triaxial test data}, volume = {161}, year = {2025}, }
• (2025). Graph attention neural network for advancing medical imaging by enhancing segmentation and classification. EAAI, 161, 112372. (www)

  @article{ , abstract = {Brain tumor classification and segmentation using Magnetic Resonance Imaging (MRI) scans remain challenging due to limited spatial relationships and the inability of conventional models, such as Convolutional Neural Networks (CNNs), to effectively capture structural dependencies. To address these limitations, this study introduces GANN-Med (Graph Attention Neural Network for Medical Imaging), a novel framework designed to improve the accuracy and reliability of brain tumor detection, segmentation, and classification. The framework utilizes the publicly available Brain Tumor MRI dataset and applies preprocessing followed by wavelet transform to extract both spatial and frequency features, generating wavelet-based node embeddings that represent fine textures and overall tumor structure. These features are modeled as a graph and processed through a U-Net (U-shaped Convolutional Network) architecture integrated with Graph Attention Networks (GATs), enabling adaptive attention to be directed to critical tumor regions. The segmented tumor regions are then classified using GraphSAGE (Graph Sample and Aggregation) with a pooling aggregator, which supports inductive neighborhood learning while preserving topological consistency. Unlike existing models such as mResU-Net (Modified Residual U-Net) and LG-GNN (Local-Global Graph Neural Network), GANN-Med uniquely combines wavelet-based multi-resolution graph embeddings with attention mechanisms in a unified architecture. Experimental results demonstrate a significant performance boost, including an 18.47 % increase in Dice Similarity Coefficient (DSC), notable improvements in Jaccard Index and sensitivity, a 21.04 % reduction in false positives, and an overall classification accuracy of 93.2 %, with a balanced accuracy of 91.8 % across glioma, meningioma, pituitary, and no tumor classes. Additionally, the model achieves minimal training and validation losses of 43.2 % and 41.6 %, respectively, highlighting its potential to minimize misdiagnosis and contribute to more accurate, early-stage clinical decision-making in neuro-oncology.}, archive = {J_EAAI}, author = {Meshari D. Alanazi and Khaled Kaaniche and Mohammed Albekairi and Turki M. Alanazi and Munid Alanazi and Ghulam Abbas}, doi = {10.1016/j.engappai.2025.112372}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112372}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph attention neural network for advancing medical imaging by enhancing segmentation and classification}, volume = {161}, year = {2025}, }
• (2025). Fuzzy scheduling in distributed heterogeneous printed circuit board assembly lines: Feedback assisted neighborhood-based search coupling with rapid evaluations. EAAI, 161, 112353. (www)

  @article{ , abstract = {Printed Circuit Board (PCB) production involves multiple heterogeneous and highly automated assembly lines running concurrently. These assembly lines involve complex processes with uncertain and fuzzy production times, presenting significant challenges for scheduling. Moreover, PCB production is notably affected by carryover sequence-dependent setup times (CSDST) and the need for product grouping, further complicating the scheduling process. To address these challenges, this paper formulates the scheduling problem as a fuzzy distributed heterogeneous flowshop group scheduling problem with carryover sequence-dependent setup times (FDHFGSP/CSDST). A novel and effective feedback-assisted population-based neighborhood search (FPBNS) algorithm, coupling with rapid evaluation of neighboring solutions, is proposed to solve this problem. The feedback strategy operates on two main levels: evaluating population quality and adjusting the search process. Additionally, an integrated heuristic method is employed to generate a high-quality and diverse initial population. To further enhance solution quality, a collaborative operation is designed to improve interaction between solutions. This comprehensive algorithm significantly enhances its adaptability in PCB manufacturing, providing an efficient solution to the complex scheduling challenges of real-world production environments. Comprehensive and systematic experiments were conducted to evaluate the effectiveness of the proposed algorithm and related methods, demonstrating its superior performance in solving the complex scheduling problem in PCB manufacturing.}, archive = {J_EAAI}, author = {Zhen-duo Han and Biao Zhang and Chao Lu and Lei-lei Meng and Wen-qiang Zou}, doi = {10.1016/j.engappai.2025.112353}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112353}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy scheduling in distributed heterogeneous printed circuit board assembly lines: Feedback assisted neighborhood-based search coupling with rapid evaluations}, volume = {161}, year = {2025}, }
• (2025). A high-precision acupoint recognition and localization model for acupuncture robot end-effectors. EAAI, 161, 112348. (www)

  @article{ , abstract = {Accurate acupoint recognition is fundamental to therapeutic efficacy of acupuncture robots, presenting a crucial challenge in medical robotics: high-precision, real-time localization of sparse features on deformable skin surfaces. To address this, we propose the Acupoint Recognition Network (ACU-Net) for reliable acupoint identification to guide robotic acupuncture. From an artificial intelligence (AI) perspective, the core contribution of ACU-Net lies in its novel modules. The Convolutional Dilatation with Residual and Split (CDRS) module uses dilated convolution to expand the receptive field and overcomes the sparsity of acupoint features by capturing both anatomical contexts and subtle local gradients. The Cross-Scale Feature Fusion (CSFF) architecture integrates spatial and semantic information more effectively than standard Feature Pyramid Network (FPN). Formulated as a keypoint regression task, the network is jointly optimized using a weighted Object Keypoint Similarity (OKS)-based loss function. Evaluated on a physician-annotated, large-scale chest acupoint dataset, ACU-Net has achieved a mean Average Precision (AP 50-95 (B)) of 94.1 % for acupoint bounding boxes and 99.5 % for acupoint points (AP 50-95 (P)), outperforming mainstream Convolutional Neural Networks (CNNs) and Transformer models. Its average prediction error (0.282 cm) on volunteers represents a 30.9 % reduction compared to the second-best model of Real-Time Multi-person Pose Estimation (RTMPose). Robustness test on an independent forearm dataset confirmed its precision. Attention visualization showed strong focus on acupoint regions. Testing on chest acupoint under adverse imaging conditions of low illumination and information loss demonstrated its reliability with zero missed detection. This work significantly advances the development of acupuncture robotics for intelligent diagnosis and treatment.}, archive = {J_EAAI}, author = {Ailing Tan and Hao Mu and Wei Ma and Yu'e Lv and Haoyu Wang and Yong Zhao}, doi = {10.1016/j.engappai.2025.112348}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112348}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A high-precision acupoint recognition and localization model for acupuncture robot end-effectors}, volume = {161}, year = {2025}, }
• (2025). Eviformer: An uncertainty fault diagnosis framework guided by evidential deep learning. EAAI, 161, 112328. (www)

  @article{ , abstract = {Unpredictable signals are commonly encountered during equipment operation, and existing deep learning-based fault diagnosis methods often fail to accurately evaluate the uncertainty of diagnostic results, limiting the model's capacity to respond to unexpected signals. To address this challenge, this study introduces a modified Transformer model based on deep evidential learning—Eviformer. The proposed model first employs a Swin-Transformer (ST)-based distribution projector, which preserves the advantages of ST in extracting features from vibration signals and simultaneously projects these signals directly into a Dirichlet distribution with second-order probabilities. Furthermore, by incorporating novel evidence correction terms and a constraint factor to reconstruct the evidence constraint loss, more precise uncertainty quantification in diagnostic predictions is achieved. Using a gear-bearing vibration dataset, comparative experiments were conducted across various scenarios, including out-of-distribution gear faults, faults in unmonitored components, noise interference, and variable speed conditions. The results demonstrate that the proposed method can promptly issue uncertainty-based warnings when encountering vibration signals that significantly differ from the training set distribution, thereby offering essential support for maintenance decisions.}, archive = {J_EAAI}, author = {Jingjie Luo and Fucai Li and Xiaolei Xu and Wenqiang Zhao and Dongqing Zhang}, doi = {10.1016/j.engappai.2025.112328}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112328}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Eviformer: An uncertainty fault diagnosis framework guided by evidential deep learning}, volume = {161}, year = {2025}, }
• (2025). Multi-scale spatiotemporal feature fusion for super-resolution video reconstruction in dynamic scenes. EAAI, 161, 112327. (www)

  @article{ , abstract = {High-quality geospatial video reconstruction is vital for applications such as urban surveillance and disaster assessment. However, existing methods often suffer from motion blur and visual artifacts due to hardware limitations and dynamic environments with rapid motion, frequent background changes, and lighting variations. To address these challenges, we propose a novel video super-resolution method called Multi-Scale Spatiotemporal Feature Fusion for Video Super-Resolution. This approach leverages artificial intelligence techniques to extract rich spatiotemporal features using a multi-scale deformable convolutional pyramid with parallel branches. It integrates contextual residual learning and optical flow estimation to capture spatial details and compensate for missing temporal information. A hybrid spatiotemporal attention mechanism adaptively fuses key features across frames through dynamic weighting, suppressing background interference and enhancing target-region representation. The framework includes an explicit motion compensation module for precise frame alignment, followed by upsampling to generate high-resolution frames with improved perceptual quality. A multi-task joint loss function optimizes structural fidelity, motion consistency, and visual realism. Experimental results on the University of Alberta Detection and Tracking dataset and Video for Super-Resolution benchmark show the method achieves a peak signal-to-noise ratio of 30.62 dB and a structural similarity index of 0.92, surpassing state-of-the-art methods by up to 4.28 dB and improving perceptual metrics such as Learned Perceptual Image Patch Similarity and Flip Loss in Perception. Ablation studies confirm each module's contribution. The model runs efficiently at 24.8 ms per frame, demonstrating potential for real-time applications. This approach offers a robust and efficient solution for high-resolution geospatial video reconstruction.}, archive = {J_EAAI}, author = {mengqin Yang and yanhui Wang and yang Yang and chunhua Peng}, doi = {10.1016/j.engappai.2025.112327}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112327}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale spatiotemporal feature fusion for super-resolution video reconstruction in dynamic scenes}, volume = {161}, year = {2025}, }
• (2025). Explainable framework based on integrated sensor fusion and deep learning to estimate stability in high-speed micromilling of thin-walled titanium alloy. EAAI, 161, 112326. (www)

  @article{ , abstract = {A novel deep learning (DL) framework based on multi-sensor fusion based on vibration and audio signals with transfer learning has been developed and validated for chatter onset prediction in micromilling of thin-walled titanium (TC4) alloys. The DL framework has been trained and validated with individual sensor data, labelled using machined surface micrographs and cutting tool condition. A total of 879,840 spectrograms has been generated for training via signal segmentation and augmentation methods that yielded a validation accuracy of 89 % during training phase. The trained DL model has been tested on the unlabelled fused signal data to make the DL model adapt to the unseen patterns in the signals recorded during real-time machining. The proposed DL framework based on fused signals outperformed individual sensor-based detection methods by demonstrating an increment of 33 % in accuracy, with a mean absolute error of 0.027416 and R 2 of 0.987. Ablation studies confirm that fusing signals enhances the DL model's performance, yielding a 21 % and 29 % improvement in precision and recall, respectively. The optimal fusion strategy reduced chatter misclassification, with one instance of misclassification. The proposed framework presents a robust, explainable DL approach that can be readily used for adaptive control of cutting conditions in different manufacturing processes. The industry readiness of the DL model has been demonstrated by testing the pretrained model on turning process. The DL model shows 95.31 % prediction accuracy with only three misclassifications out of sixty-four test conditions.}, archive = {J_EAAI}, author = {Sethurao Gururaja and Kundan K. Singh}, doi = {10.1016/j.engappai.2025.112326}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112326}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable framework based on integrated sensor fusion and deep learning to estimate stability in high-speed micromilling of thin-walled titanium alloy}, volume = {161}, year = {2025}, }
• (2025). Liquid metal microfluidic cooling system for high-efficiency thermal management via learning-based genetic algorithm. EAAI, 161, 112324. (www)

  @article{ , abstract = {High heat flux density is a critical factor that limits the performance and reliability of miniaturized, high-power microelectronic systems. This study proposes a liquid metal (LM)-based microfluidic cooling system optimized through a data-driven computational framework based on an enhanced Genetic Algorithm (LC-GA), aiming to deliver an efficient thermal management solution for high-density integrated systems. By integrating LM near-junction cooling with microchannel heat dissipation in a silicon substrate, we developed a heterogeneous three-dimensional interconnect cooling architecture capable of optimizing thermal performance through algorithm-guided parameter tuning. To validate the proposed method, four distinct microchannel configurations were designed, fabricated, and experimentally tested. LM was introduced into the channels to conduct both experimental cooling tests and thermal performance simulations on a simulated heat source. The results demonstrate that this LM-based microfluidic cooling system, optimized through computational parameter determination, can effectively dissipate heat from chips with power consumption up to 800 W while maintaining stable thermal performance. Additionally, a response surface methodology combined with enhanced LC-GA was utilized for multi-factor sensitivity analysis and multi-objective optimization, enabling automatic determination of optimal design and operating parameters to balance thermal resistance and pressure drop. The optimized configuration reduced the maximum chip temperature to approximately 357.54 K, lowered the system pressure requirement, and improved the Performance Evaluation Criterion (PEC) to 2.327. This work provides a data-driven optimization approach that supports the development of high-performance integrated microsystems through algorithm-assisted thermal design.}, archive = {J_EAAI}, author = {Yucheng Wang and Antong Bi and Kaiyu Chen and Shenxin Yu and Wanping Gao and Wenyi Zhang and Yuwan Wu and Zhiqiang Li and Shaoxi Wang}, doi = {10.1016/j.engappai.2025.112324}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112324}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Liquid metal microfluidic cooling system for high-efficiency thermal management via learning-based genetic algorithm}, volume = {161}, year = {2025}, }
• (2025). HoloFluoNet: Live cell imaging intelligence based on fused holography and fluorescence for virtual staining, cell segmentation, classification, and viability analysis. EAAI, 161, 112323. (www)

  @article{ , abstract = {Fluorescence microscopy enables detailed visualization of subcellular structures but is limited by phototoxicity, photobleaching, and labor-intensive labeling. In contrast, digital holographic microscopy (DHM) offers label-free, quantitative phase imaging but lacks biochemical specificity. To integrate the strengths of both modalities, we propose HoloFluoNet, a deep learning-based framework that generates virtual fluorescence information from phase images acquired by DHM. Using a dual-mode imaging system, we simultaneously captured phase and fluorescence images of live cancer cells. The fluorescence data were used to construct biologically grounded supervision signals, including nuclei masks and multi-class cell viability labels. From a single-phase image, HoloFluoNet predicts a virtual nuclei mask, a distance map for boundary refinement, and a viability mask. The architecture incorporates multi-scale feature extractor, and attention mechanisms, optimized with novel inclusion and exclusion loss functions. Post-processing using the watershed algorithm ensures accurate segmentation of overlapping cells. The final registered image provides label-free virtual staining for nuclei localization, viability assessment, and class-specific morphological profiling. HoloFluoNet achieved a Dice score of 0.8685 and an Aggregated Jaccard Index (AJI) of 0.7883, outperforming conventional deep learning models such as U-Net, Fully Convolutional Network (FCN), Pyramid Scene Parsing Network (PSPNet), and DeepLab v3+. These improvements were statistically validated using one-way analysis of variance (p < 0.01). Ablation experiments confirmed the complementary roles of architectural modules and novel loss functions, while robustness tests under noisy and low-quality conditions revealed high stability to low contrast and moderate noise. With an inference speed of 0.123 s per image, the model enables real-time cellular analysis. By bridging structural and molecular imaging, HoloFluoNet provides an efficient, label-free alternative to fluorescence microscopy, with promising applications in artificial intelligence (AI)-assisted drug screening, cancer research, and live-cell monitoring.}, archive = {J_EAAI}, author = {Seonghwan Park and Jaeseong Lee and Jaewoo Park and Inkyu Moon}, doi = {10.1016/j.engappai.2025.112323}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112323}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {HoloFluoNet: Live cell imaging intelligence based on fused holography and fluorescence for virtual staining, cell segmentation, classification, and viability analysis}, volume = {161}, year = {2025}, }
• (2025). Method for ensuring quasi-maximum stability of a system with interval deviations by a robust controller parametric synthesis. EAAI, 161, 112320. (www)

  @article{ , abstract = {It is known that ensuring reliable stability of automatic control systems with interval and stochastic parameter deviations is critical for aviation and other critical areas. Therefore, this article proposes a method for synthesizing Proportional-Integral-Derivative controller parameters based on the coefficient approach to stability analysis (through the characteristic polynomial neighboring coefficients ratios) with an extension to interval uncertainties and the stability quasi-maximum degree integration into the multi-criteria optimization Non-Dominated Sorting Genetic Algorithm II. Scientific contribution and novelty lie in combining analytical sufficient stability conditions (including verification by Kharitonov polynomials and dominant polynomial analysis) with multi-objective optimization taking into account parameter errors and energy costs, which allows obtaining stable and economical Proportional-Integral-Derivative settings for discrete two-channel systems. The results are confirmed by numerical experiments on a helicopter turboshaft engine two-channel control system model with real on-board data. It is shown that the optimal parameter vector provides overshoot of ≈1.43 % and transient process time of ≈1.12 s and maintains stability with local changes in coefficients (±3 … 7 %) and in Monte-Carlo tests. At the same time, a comparative analysis with traditional and neural network controllers showed the proposed approach's advantage in key metrics.}, archive = {J_EAAI}, author = {Denys Baranovskyi and Serhii Vladov and Valerii Sokurenko and Oleksandr Muzychuk and Victoria Vysotska and Maryna Bulakh}, doi = {10.1016/j.engappai.2025.112320}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112320}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Method for ensuring quasi-maximum stability of a system with interval deviations by a robust controller parametric synthesis}, volume = {161}, year = {2025}, }
• (2025). A robust deep learning model selection with data augmentation for automatic detection of tessellated fundus images and explainable artificial intelligence based interpretation. EAAI, 161, 112316. (www)

  @article{ , abstract = {A robust deep learning system for automatically classifying retinal fundus images into two classes—normal and tessellated—is presented in this study. Visual Geometry Group – 16 is used as the base model, taking advantage of transfer learning to develop an efficient framework for fundus image classification. The approach uses nine different model architectures and makes use of a dataset of 352 fundus images that was increased to 4865 samples using sophisticated data augmentation techniques. Of these, Model_8 performed the best, achieving a loss of 0.129 % and an impressive accuracy of 99.39 %. The suggested approach ensures higher performance and dependability by combining rigorous data augmentation, efficient preprocessing, and model fine-tuning techniques. Furthermore, Explainable Artificial Intelligence was used to improve the interpretability of the model and visualize important aspects such as features or imposed pathologies in fundus images more clearly. The study offers promising support to ophthalmologists by offering precise automated diagnoses for the early identification and treatment of retinal disorders.}, archive = {J_EAAI}, author = {Kachi Anvesh and Shanmugasundaram Hariharan and Bharati M. Reshmi and Qiang Xu and Joan Lu and Vinay Kukreja and Murugaperumal Krishnamoorthy}, doi = {10.1016/j.engappai.2025.112316}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112316}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust deep learning model selection with data augmentation for automatic detection of tessellated fundus images and explainable artificial intelligence based interpretation}, volume = {161}, year = {2025}, }
• (2025). Camera-invariance correlation learning and inter-domain-specific distinct representation for person re-identification. EAAI, 161, 112314. (www)

  @article{ , abstract = {Unsupervised domain adaptation (UDA) in person re-identification (ReID) experiences performance degradation due to two primary factors: (1) the information changes across cameras among intra-domain identity samples within the source and target domains; and (2) the inter-domain gap when applied to an unexplored image domain, which makes it a challenging task. The majority of current methodologies predominantly address the inter-domain gap while overlooking the impact of the intra-domain issue. To overcome this limitation, this paper proposes an effective invariance representation ReID (IR-ReID) network that encompasses camera-invariance correlation learning (CICL) and Transformer-based, inter-domain-specific discriminative representation (IDSR), supported by curriculum learning, with the aim of concurrently addressing both of these pivotal issues. Specifically, a CICL module is designed to explore intra-domain distinct features of an individual captured by multiple cameras across the source and target domains. Secondly, an IDSR module is proposed to address the inter-domain gap issue. Thirdly, a curriculum learning-driven strategy is introduced for sample training, systematically guiding our network in learning samples progressively, ranging from easy to difficult instances. Meanwhile, a cluster-based constraint loss is presented to further optimize the clustering results. Finally, extensive experiments are conducted on three well-known datasets, which demonstrate the effectiveness and superiority of our proposed approach in comparison to the prevailing state-of-the-art methodologies.}, archive = {J_EAAI}, author = {Shangdong Zhu and Yunzhou Zhang and Peng Duan}, doi = {10.1016/j.engappai.2025.112314}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112314}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Camera-invariance correlation learning and inter-domain-specific distinct representation for person re-identification}, volume = {161}, year = {2025}, }
• (2025). Applications of machine vision technology for conveyor belt deviation detection: A review and roadmap. EAAI, 161, 112312. (www)

  @article{ , abstract = {Conveyor belt deviation is a frequent challenge in product transportation filed, and failure to promptly detect and rectify this anomaly not only significantly reduces transport efficiency but also poses a risk of serious safety accidents, leading to enormous economic losses. Traditional contact-based deviation detection technologies, with their inherent limitations of high costs and complicated maintenance, have struggled to meet the practical demands of long-distance conveyor belt inspection. In this context, non-contact machine vision technology has emerged as a prominent solution in the field of conveyor belt deviation detection, thanks to its notable advantages of a simple hardware structure and round-the-clock operational capability. Recently, with the rapid development of artificial intelligence theories, this research field has accumulated a series of effective solutions that have been proven through machine vision practical applications. This paper delves into the technical principles of the existing solutions, systematically summarizes them and objectively evaluates their strengths and weaknesses in practical applications. Based on this foundation, this paper also provides an insight on the future development trends of intelligent monitoring for conveyor belt deviation, aiming to offer valuable reference and guidance to technicians in related fields.}, archive = {J_EAAI}, author = {Jiaming Han and Ting Fang and Wensheng Liu and Chenxiao Zhang and Molin Zhu and Jibin Xu and Jie Ji and Xianhua He and Zhang Wang and Min Tang and Chong Dong and Long Ma and Xinlong Yang}, doi = {10.1016/j.engappai.2025.112312}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112312}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applications of machine vision technology for conveyor belt deviation detection: A review and roadmap}, volume = {161}, year = {2025}, }
• (2025). Synthetic aperture radar target recognition with limited training data based on frequency-domain-assisted dual-stream attention hierarchical deformable convolutional networks. EAAI, 161, 112309. (www)

  @article{ , abstract = {In the field of synthetic aperture radar (SAR) automatic target recognition (ATR), the recognition performance of deep learning methods is often constrained by sample scarcity. To address this issue, this paper proposes a frequency-domain-assisted dual-stream attention hierarchical deformable convolutional network (DAHDF-Net). A frequency-domain-assisted dataset construction method is designed to extend the representational diversity of the sample feature space. Then, a parallel input architecture of frequency-space dual-stream features is constructed, and the adaptive dynamic fusion of frequency-domain stream features with spatial-domain stream features is realized by using a channel attention module in the frequency-domain stream. To enhance the expressiveness of the network on the basis of modulated deformation convolution, a hierarchical residual (H-residual) module is designed. Moreover, a triple hybrid loss is proposed to address the issues of high deformation, which makes classification difficult, and the fuzzy decision boundaries between categories. Experiments were conducted on the moving and stationary target acquisition and recognition (MSTAR) dataset and the FUSAR-Ship dataset. Among them, DAHDF-Net achieves recognition accuracies of 99.75 % and 97.48 % on all training sets and 10-way 25-shot under standard operating condition (SOC) of MSTAR, which demonstrated superior recognition accuracy compared to current state-of-the-art methods.}, archive = {J_EAAI}, author = {Shanliang Yuan and Hongyuan Gao and Xiaoyuan Gu and Jingya Ma}, doi = {10.1016/j.engappai.2025.112309}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112309}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic aperture radar target recognition with limited training data based on frequency-domain-assisted dual-stream attention hierarchical deformable convolutional networks}, volume = {161}, year = {2025}, }
• (2025). Multi-scale distance similarity entropy: A novel complexity measurement for gearbox fault diagnosis. EAAI, 161, 112308. (www)

  @article{ , abstract = {Entropy-based methods have gained increasing attention in fault diagnosis due to their capability in quantifying the intrinsic complexity and dynamic behavior of time series signals. Vibration signals from gearboxes are typically nonlinear, nonstationary, and transient, often spanning multiple time scales, thus requiring robust multi-scale analysis for accurate condition assessment. However, existing multi-scale entropy methods, prone to missing local variations and vulnerable to noise, often yield incomplete features and low diagnostic accuracy with high computational costs. To overcome these challenges, this paper proposes a novel complexity measurement method termed multi-scale distance similarity entropy, which integrates distance similarity entropy with a multi-scale coarse-graining process. By computing element-wise distances and mapping them through a Gaussian kernel function, multi-scale distance similarity entropy captures subtle nonlinear variations in vibration signals while effectively suppressing noise interference. It estimates the similarity distribution between adjacent vectors, enabling sensitive tracking of dynamic complexity changes in mechanical vibration data. Experimental validation on two gearbox datasets encompassing both single and compound faults demonstrates that multi-scale distance similarity entropy achieves superior classification accuracies of 98.63 % and 97.14 %, respectively. The method also shows strong robustness under random impulse noise and low signal-to-noise ratio conditions, along with high computational efficiency, particularly in small-sample scenarios. These characteristics make it highly suitable for online real-time fault diagnosis in complex and noise contaminated industrial environments involving diverse and compound fault patterns.}, archive = {J_EAAI}, author = {Tao Wang and Shin Yee Khoo and Zhi Chao Ong and Pei Yi Siow and Teng Wang}, doi = {10.1016/j.engappai.2025.112308}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112308}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale distance similarity entropy: A novel complexity measurement for gearbox fault diagnosis}, volume = {161}, year = {2025}, }
• (2025). JSSM-DTA: Joint sequence-structure modeling with multi-scale transformers for explainable drug-target affinity prediction. EAAI, 161, 112303. (www)

  @article{ , abstract = {Drug-Target Affinity (DTA) prediction is a pivotal task in drug discovery, aiming to quantify the strength of interactions between drug molecules and their biological targets, such as proteins. Despite significant advancements in deep learning-based methods, several challenges persist. Scale mismatches obstruct functional group-specific representations, leading to suboptimal feature extraction and loss of critical interaction patterns. Additionally, the inability to fully exploit molecular graph structures weakens hierarchical dependency modeling, while ineffective feature fusion hampers the integration of heterogeneous molecular representations, restricting the capture of intricate drug-target interaction dynamics. To address these challenges, this work introduces JSSM-DTA, a novel Joint Sequence-Structure Modeling framework that integrates Multi-Scale Transformers with intermodal feature attention to enhance DTA prediction. JSSM-DTA formulates a unified representation space through the Adaptive Convolutional Transformer (ACT) and the Multi-Scale Diffusion Transformer (MSDT), where ACT refines sequential feature extraction by capturing intricate local dependencies and long-range contextual relationships, while MSDT preserves hierarchical learning, optimizing both global topological organization and fine-grained interactions. Additionally, the Factorized Inter-layer Interaction Module constructs joint embeddings by seamlessly integrating heterogeneous representations, enabling robust feature fusion. The proposed model was evaluated on Davis, KIBA, Metz and BindingDB datasets, demonstrating superior predictive accuracy over state-of-the-art methods. Explainability analysis reveals key molecular substructures and binding interactions, enhancing transparency in decision-making and establishing JSSM-DTA as a benchmark for both accuracy and interpretability in DTA prediction.}, archive = {J_EAAI}, author = {Uma E. and Mala T.}, doi = {10.1016/j.engappai.2025.112303}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112303}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {JSSM-DTA: Joint sequence-structure modeling with multi-scale transformers for explainable drug-target affinity prediction}, volume = {161}, year = {2025}, }
• (2025). Cyclic translations between pathomics and genomics improve automatic cancer diagnosis from whole slide images. EAAI, 161, 112302. (www)

  @article{ , abstract = {Incorporating genomic characterization into histopathological image modeling brings substantial value to enhancing diagnostic accuracy and supporting the development of targeted and effective treatment strategies. However, prevailing multi-modal integration methods often assume the availability of both pathomics and genomics data in both training and testing phases, overlooking the challenge of data absence due to prohibitive costs. In this paper, we propose a multi-modal cyclic feature generation network (MCFGN) that facilitates cyclic translations between pathomics and genomics to acquire a unified representation of multi-modal data. This approach enables the use of pathological images alone as input to generate joint representations during the testing phase. First, we utilize a general-purpose, self-supervised vision encoder to embed histological image patches as distinctive visual tokens. Next, we hierarchically aggregate patch-level tokens to region-level and slide-level, generating improved whole slide image (WSI) representations. We build self-supervised Masked Autoencoders (MAE) to initialize the hierarchical aggregator. Finally, to incorporate genomic characterization into the learning process, we develop a novel cross-modal cyclic feature generation module to create an intermediate joint representation of pathological and genetic features for patient diagnosis. Evaluations have been conducted on two public datasets from The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) and Non-Small Cell Lung Cancer (TCGA-NSCLC) for various diagnostic tasks, including cancer subtyping and biomarker status prediction. Experiments indicate that our MCFGN model improves predictive performance in cancer diagnosis using histological slides, yielding an 8.7% improvement in area under the curve (AUC) for the cancer subtyping task and a 14.1% gain for biomarker prediction.}, archive = {J_EAAI}, author = {Xinyu Hao and Hongming Xu and Xiaofeng Wang and Tong Wang and Timo Hamalainen and Fengyu Cong}, doi = {10.1016/j.engappai.2025.112302}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112302}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cyclic translations between pathomics and genomics improve automatic cancer diagnosis from whole slide images}, volume = {161}, year = {2025}, }
• (2025). A multimodal industrial anomaly detection method based on mask training and teacher–student joint memory. EAAI, 161, 112299. (www)

  @article{ , abstract = {In recent years, the teacher–student framework has been applied to both single-modality detection and multimodal detection, which realizes anomaly detection based on the feature difference between the teacher model and the student model. However, current multimodal teacher–student models use the same teacher model to extract two-dimensional (2D) image and three-dimensional (3D) point cloud features. The point cloud features extracted by the teacher model pre-trained on images are not the optimal feature representation. To further improve the performance of the teacher–student framework on the multimodal anomaly detection task, this paper proposes M ultimodal T eacher- S tudent J oint M emory ( MTSJM ). MTSJM constructs a teacher–student joint memory bank for each modality, the feature distance between the test sample and the memory bank is used as the anomaly indicator. This distance reflects the feature differences between the test sample and the normal sample at multiple levels, including the teacher–teacher, teacher–student, and student–student levels. Then, this paper proposes a mask-based student model training method. While ensuring that the student learns the feature of normal regions, mask training increases the feature difference of non-normal regions between the student and the teacher. On the MVTec 3D Anomaly Detection (MVTec 3D-AD) dataset, the proposed MTSJM achieves effective anomaly detection performance, reaching 95.7% mean Image-level Area Under the Receiver Operator Curve (I-AUROC) and 97.2% mean Area Under the Per-Region Overlap (AUPRO). In addition, MTSJM achieves 99.3% I-AUROC and 99.6% Pixel-level AUROC (P-AUROC) on a real-world vehicle stamping part task, which further illustrates the applicability of MTSJM on the multimodal anomaly detection task.}, archive = {J_EAAI}, author = {Yi Liu and Changsheng Zhang and Xingjun Dong and Yufei Yang}, doi = {10.1016/j.engappai.2025.112299}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112299}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multimodal industrial anomaly detection method based on mask training and teacher–student joint memory}, volume = {161}, year = {2025}, }
• (2025). Fuse MetaFormer with convolutional neural networks for three-dimensional model classification. EAAI, 161, 112298. (www)

  @article{ , abstract = {Multimedia technology is widely applied to artificial intelligence and it is key to the performance of artificial intelligence systems, such as computer-aided design system, virtual reality system, the augmented reality system, medical image processing system, and game development system. With the development of multimedia technology, the number of three-dimensional (3D) models in network or database is becoming larger and larger. It is important to classify 3D models. In order to improve accuracy of three-dimensional model classification, a method of 3D model classification fusing MetaFormer with CNN (Convolutional Neural Networks) is proposed. 3D model is projected into two-dimensional (2D) views through the fixed-view projection, and representative views are selected by the clustering algorithm. Points sampled randomly from representative view are used to calculate its shape descriptors. View feature is extracted from representative view by MetaFormer. Shape feature is extracted from shape descriptors of representative view by Convolutional Neural Networks. View feature and shape feature of representative view are fused. At the same time, majority voting algorithm is used to determine category of 3D model based on the fusion of view features and shape features. Experiments are conducted on ModelNet10 dataset. Experimental results show that the proposed method achieves better results than others.}, archive = {J_EAAI}, author = {Xueyao Gao and Haomin Wu and Chunxiang Zhang and Yongzeng Xue}, doi = {10.1016/j.engappai.2025.112298}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112298}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuse MetaFormer with convolutional neural networks for three-dimensional model classification}, volume = {161}, year = {2025}, }
• (2025). Intelligent design of broadband metasurface based on spectrum prediction neural network and transition simulated annealing algorithm. EAAI, 161, 112297. (www)

  @article{ , abstract = {Recently, metasurfaces have demonstrated considerable potential for application across various fields, including wireless communication, Internet of Things, holographic imaging, and radar systems. With the rapid development of artificial intelligence and its application in other disciplines, metasurface design based on artificial intelligence techniques, particularly deep learning, has attracted widespread attention. In this paper, we propose a novel scheme that integrates deep neural networks with an enhanced simulated annealing algorithm to achieve intelligent design of broadband metasurfaces. Firstly, we construct and train a neural network capable of accurately predicting the amplitude and phase responses of an isotropic metasurface structure, achieving a Mean Absolute Error of 1.36° on the test set. Subsequently, we develop a novel simulated annealing algorithm that is integrated with the prediction neural network to form an intelligent optimization framework. This approach can effectively search for the solution space to accurately achieve arbitrary target phase response, with a very low Mean Squared Error calculated on the test set. Furthermore, we validate the effectiveness of the proposed model and algorithm through a practical application that demonstrates broadband dispersion characteristics within the frequency range of 9–11 Gigahertz (GHz). Additionally, the prediction model can employ transfer learning strategy to adapt to various operating frequency bands, significantly enhancing its generalization capabilities. The proposed intelligent design methodology introduces a novel pathway for rapidly designing metasurface in complex application scenarios, serving as a valuable reference for the engineering design of devices in electronics, optics, and wireless communication.}, archive = {J_EAAI}, author = {Yin Zhang and Yuxin Dai and Yiyu Fan and Jun Yu}, doi = {10.1016/j.engappai.2025.112297}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112297}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent design of broadband metasurface based on spectrum prediction neural network and transition simulated annealing algorithm}, volume = {161}, year = {2025}, }
• (2025). Industrial fabric defect-generative adversarial network (IFD-GAN): High-fidelity fabric cross-scale defect samples synthesis method for enhancing automated recognition performance. EAAI, 161, 112296. (www)

  @article{ , abstract = {—High-quality defect samples and datasets are the cornerstone for enhancing the performance of deep learning detection algorithms for fabric defects. Constraints, such as the complexity of fabric textures, the diversity and sporadic occurrence of defects, the insufficient scale, and the significant class imbalance of datasets, impede improvements in detection accuracy and generalization of detection models, thereby limiting industrial application. To address the issues above, a two-stage generative model for fabric defect images called Industrial Fabric Defect-Generative Adversarial Network (IFD-GAN), based on contrastive learning mutual information mechanism, was introduced. The Pixel-level Defect-Background Stripper (DBS) was devised for precise localization, and the Longitude-Latitude Self-Attention Mechanism (LLSA) was proposed to efficiently focus on defect foreground details, facilitating decoupling and efficient extraction of defect features. The combination of the two, along with the structural similarity loss function, collectively regulates the coordination and consistency between defect and background textures. An industrial-grade fabric defect dataset was collected for IFD-GAN training and generative testing. The model's effectiveness in generating defect images was evaluated based on dimensions such as fidelity and diversity. IFD-GAN is used to enhance this dataset, and the system compared the performance of advanced object detection models trained on the datasets before and after the enhancement. Extensive experimental results show that IFD-GAN can accomplish high-fidelity generation of high-resolution cross-scale fabric defect images in an unsupervised environment, significantly contributing to the creation of more balanced and diverse large-scale industrial fabric defect datasets and the enhancement of defect detection networks in precision, robustness, and generalizability.}, archive = {J_EAAI}, author = {Shun Xu and Sheng Cheng and Shuyang Jin and Xudong Hu and Weitao Wu and Zhong Xiang}, doi = {10.1016/j.engappai.2025.112296}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112296}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Industrial fabric defect-generative adversarial network (IFD-GAN): High-fidelity fabric cross-scale defect samples synthesis method for enhancing automated recognition performance}, volume = {161}, year = {2025}, }
• (2025). Sample-efficient planning-control framework for autonomous underwater vehicle docking using lightweight cross Q-learning. EAAI, 161, 112293. (www)

  @article{ , abstract = {Autonomous Underwater Vehicles (AUVs) are essential for long-duration missions in underwater sensor networks, but their operation is constrained by limited onboard energy and low-bandwidth communication. Docking stations are deployed to support energy replenishment and data transfer, requiring AUVs to autonomously plan return trajectories and achieve precise docking. Traditional motion planning and control methods for AUV docking often decouple global motion planning from local control or rely on accurate dynamics models, while existing deep reinforcement learning (DRL) approaches frequently neglect sample efficiency. This work presents a novel DRL framework that tightly integrates motion planning and control while emphasizing sample efficiency to enable reliable and efficient AUV docking. The docking task is formulated as an Markov Decision Process (MDP), in which a structured reward function is defined to guide the agent towards goal-directed behavior in an interpretable and sample-efficient manner. A lightweight Cross Q-Learning (CrossQ-Lite) algorithm is developed by removing target networks and incorporating batch normalization, achieving stable learning with approximately 92.8% fewer parameters while maintaining comparable performance. Simulation results demonstrate that, under the proposed MDP formulation, CrossQ-Lite achieves a minimum of 1 . 39 × higher sample efficiency and improved training stability compared to state-of-the-art DRL baselines. The trained model reliably completes AUV docking tasks, providing a practical and high-performance solution for autonomous underwater sensor networks.}, archive = {J_EAAI}, author = {Kaixin Zhang and Yuelong Xu and Minghao Zhao and Yu Jiang}, doi = {10.1016/j.engappai.2025.112293}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112293}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sample-efficient planning-control framework for autonomous underwater vehicle docking using lightweight cross Q-learning}, volume = {161}, year = {2025}, }
• (2025). Fault diagnosis of wind turbine based on dual-channel feature aggregation network with attentional mechanism. EAAI, 161, 112291. (www)

  @article{ , abstract = {Wind turbines are subjected to alternating stresses and shock loads during operation, and the collected vibration signals are nonlinear, non-stationary and contain noise, resulting in subtle fault features that are difficult to extract. In order to extract discriminative features from vibration signals under variable speed operating conditions, a dual-channel feature aggregation network (DCNet) with attention mechanism is developed in this paper. First, a Parallel Patch-Aware Convolutional Module (PPCM) is constructed to extract feature information of different scales and levels from Time–Frequency Representations (TFR). Specifically, convolutional operations are performed in both the spatial and frequency domains, endowing it with local–global capturing capabilities and efficiency. Then, to improve the network operation rate, Haar Wavelet Downsampling (HWD) is embedded into the DCNet architecture. The core idea is to reduce the spatial resolution of features through Haar wavelet transform while preserving as many discriminative features as possible. Additionally, Channel Prior Convolutional Attention (CPCA) is introduced to enable DCNet to focus on more critical feature parts by dynamically allocating channel and spatial attention weights, thereby suppressing redundant feature interference. Finally, in the Deep Feature Fusion Module (DFFM), a cross-attention mechanism is adopted for global interaction, fusing features from different channels to achieve feature enhancement. To obtain good diagnostic results under variable speed conditions, we apply a label smoothing algorithm to assist model training. The experimental results show that the proposed model has high diagnostic accuracy, can generalize effectively, and it still maintains high diagnostic accuracy under noise and variable speed working conditions.}, archive = {J_EAAI}, author = {Haiyu Guo and Xingzheng Guo and Xiaoguang Zhang and Fanfan Lu and Chuang Liang}, doi = {10.1016/j.engappai.2025.112291}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112291}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of wind turbine based on dual-channel feature aggregation network with attentional mechanism}, volume = {161}, year = {2025}, }
• (2025). A new deep learning framework for intelligent aerial monitoring of power transmission line insulators. EAAI, 161, 112290. (www)

  @article{ , abstract = {Transmission line insulators are critical components for maintaining the integrity and efficiency of power transmission lines. Monitoring the health, performance, and efficiency of this network has always been of interest since the past. Traditional and field inspection methods are often tedious, costly, and prone to human error. In order to monitor the condition of transmission lines and overcome traditional challenges, this study proposes an intelligent monitoring framework using unmanned aerial vehicle (UAV) imaging and advanced deep learning models to identify and classify insulator defects. The dataset used in this study consists of 870 original images, 600 healthy images, 140 damaged images, and 130 images of flashover insulators collected from UAV flights and online sources. To increase the diversity of the training data, a data augmentation technique was used, increasing the number of images to 1952 samples and dividing the data into 80 % training sets and 20 % test sets. While the current evaluation is performed on this dataset, real-world experiments will be considered in future studies to further validate the model. In this approach, three scenarios were considered to evaluate and detect these errors. The first scenario used an independent convolutional neural network (CNN) for feature extraction and classification, achieving an accuracy of 82.54 %. The second scenario combined CNN with transfer learning (TL) techniques to improve feature representation, achieving an accuracy of 95.99 %. Scenario Three integrated a hybrid CNN model with a long short-term memory network, achieving an average accuracy of 99.67 % in fault detection and classification. The results demonstrate the superiority and robustness of the proposed hybrid model, providing a reliable and cost-effective solution for transmission line insulator monitoring. This study highlights the significant potential of combining UAV imagery with deep learning algorithms, reassuring the audience of the effectiveness of the proposed solution.}, archive = {J_EAAI}, author = {Reza Rahimi Nejadbougar and Ebadat Ghanbari Parmehr and Alireza Afary and Samira Mavaddati}, doi = {10.1016/j.engappai.2025.112290}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112290}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new deep learning framework for intelligent aerial monitoring of power transmission line insulators}, volume = {161}, year = {2025}, }
• (2025). Spatial-temporal generative network based on deep long short-term memory autoencoder for hand skeleton data sequences reconstruction and recognition. EAAI, 161, 112289. (www)

  @article{ , abstract = {Convolutional neural networks attract the highest research focus in the developing field of Hand Gesture Recognition (HGR). Nevertheless, these approaches presented a challenging task in adapting to time-series data. In skeleton-based HGR, extracting spatial–temporal information remains a challenge. In recent times, recurrent neural networks have exhibited exceptional performance in detecting desired hand gestures by processing of varied length time-series data. Although they outperform traditional methods when huge training data is accessible, their effectiveness significantly diminishes when data availability is constrained. In this study, we introduce an unsupervised data augmentation network known as the Spatial-Temporal Generative Network (STGN), which reconstructs both the spatial and temporal information of the input sequences by leveraging a Deep Long Short-Term Memory Auto-Encoder (DLSTM-AE) network. Consequently, the DLSTM-AE combined with different Long Short-Term Memory (LSTM) network variations, forming an integrated network that can be trained end-to-end for HGR. Through experimentation conducted on the LeapGestureDB dataset (Leap Motion-based Gesture Dataset) and RIT dataset (Rochester Institute of Technology Hand Gesture Dataset), we prove that data reconstruction using STGN had a prominent effect on improving the accuracy of recognizing time-series based hand gestures. For all experiments, the best recognition results are achieved in the augmented dataset. Accuracies were improved on all tested LSTM networks from 2 to 10%. For reproducible research, the code is available at: https://github.com/AMEURsafa/STGN .}, archive = {J_EAAI}, author = {Safa Ameur and Mohamed Ali Mahjoub and Anouar Ben Khalifa}, doi = {10.1016/j.engappai.2025.112289}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112289}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial-temporal generative network based on deep long short-term memory autoencoder for hand skeleton data sequences reconstruction and recognition}, volume = {161}, year = {2025}, }
• (2025). Neural dynamic fluid reconstruction technique for four-dimensional imaging of combustion flame based on deep learning. EAAI, 161, 112288. (www)

  @article{ , abstract = {Three-dimensional optical diagnostic techniques based on the principles of tomographic imaging enable the acquisition of rich three-dimensional information in experimental flow fields through reconstruction calculations. However, for tasks involving the reconstruction of three-dimensional flow fields at high temporal resolutions, existing methods incur high computational costs, low reconstruction efficiency, and struggle to achieve high spatiotemporal resolution measurements. This paper proposes the Neural Dynamic Fluid Reconstruction Technique (NDFRT) based on deep learning. NDFRT incorporates the time dimension into the reconstruction scope to achieve the four-dimensional reconstruction of dynamic flow fields using neural networks. NDFRT has the following technical advantages: (1) ultra-high spatiotemporal reconstruction resolution; (2) good computational efficiency, with the reconstruction parameter scale only half that of traditional voxel-based methods; (3) the ability to perform three-dimensional frame prediction of dynamic fluids. We validated the proposed method using numerical simulation and experimental jet flame reconstruction and compared it with the traditional algebraic reconstruction technique (ART). Experimental results demonstrate that NDFRT outperforms traditional ART methods in terms of computational efficiency, reconstruction resolution, and reconstruction accuracy.}, archive = {J_EAAI}, author = {Fuhao Zhang and Zhiyin Ma and Can Gao and Gang Xun and Qingchun Lei and Xuesong Li}, doi = {10.1016/j.engappai.2025.112288}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112288}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural dynamic fluid reconstruction technique for four-dimensional imaging of combustion flame based on deep learning}, volume = {161}, year = {2025}, }
• (2025). BeltDiff: Diffusion-based self-labeled generation system for conveyor belt damage detection. EAAI, 161, 112287. (www)

  @article{ , abstract = {The damages of conveyor belt are abnormal occurrence of industrial production line, generally resulting in serious economic losses. While utilizing deep-learning-based methods in belt damage detection task, models training is challenging due to the prevalent scarcity of both numbers and diversity in dataset, leading to the restricted performance and limited accuracy in practical applications. To address this issue, we propose a novel self-labeled generation system based on diffusion model, for creating effective and large-scale conveyor belt detection datasets. Specifically, the training of our system is divided into three stages for progressive task decomposition and interpretable user control. Among the generation process, we focus on the textual prompt alignment and shape guidance to generate more reasonable and high-quality damaged belt images. Meanwhile, larger scale datasets are created via our method on two scenes, including base scene with limited samples, and brand new scenes with no damaged samples, which compose 9636 and 7500 images respectively. By evaluating the quality of damaged images, and the relations between generated and real images, our method demonstrates its effectiveness on generating damaged belt image pairs with annotations. Further, for verifying the validity of our generative datasets, we implement numerous experiments of 4 types of popular detection models training on different settings. The results demonstrate that our datasets support effective accuracy improvement, comparing with the base datasets and the classic augmented datasets, specifically increased by up to 14.4% in the brand new scenes with no damaged samples.}, archive = {J_EAAI}, author = {Peixian Zhuang and Yuanxiu Cai and Xi Liu and Xianchao Zheng and Fuheng Xiao and Jiangyun Li}, doi = {10.1016/j.engappai.2025.112287}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112287}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BeltDiff: Diffusion-based self-labeled generation system for conveyor belt damage detection}, volume = {161}, year = {2025}, }
• (2025). A robust deep learning based model for denoising phonocardiogram signals in clinical environments. EAAI, 161, 112286. (www)

  @article{ , abstract = {A cardiac auscultation, being a very popular first line cardiovascular diseases screening method, typically produces phonocardiogram signals contaminated by background sounds, e.g., speech and movement of other patients and doctors/nurses, phone ringing and door knocking/opening/closing, and internal physiological body noises, e.g., lung sounds, muscle contraction and motion artifacts, as it is usually done in a noisy environment and it is not possible to suppress all the internal physiological body noises. So, a quality of the recording is finally decreased. Moreover, this can strongly affect a subsequent analysis/classification of the captured recording and can also lead to its incorrect classification. In this paper, we propose a novel deep convolutional architecture that integrates the bidirectional long short-term memory layers with the transformer blocks involving multi-head attention mechanism, which is particularly designed for denoising phonocardiogram signals. The performance of the proposed artificial intelligence-based approach is evaluated through synthetic data generated by using one public phonocardiogram dataset and two public real-world hospital ambient/respiratory sound databases as well as by deploying two public and renowned real-world clinical environment datasets containing phonocardiogram signals contaminated by various environmental and physiological noises recorded during the clinical auscultations. The deployed complex evaluation strategy assures a higher robustness and generalizability of the proposed approach. As also a result of that, the proposed model outperforms the current state-of-the-art approaches and provides the significant improvements in all the assessed signal quality and noise suppression metrics and over all the involved benchmark datasets. Finally, it should be noted here that the model achieved an average estimated SNR of 14.984 dB and 11.924 dB for the PASCAL and CirCor22 dataset respectively, which represent the renowned real-world clinical environment datasets.}, archive = {J_EAAI}, author = {Maros Jakubec and Eva Lieskovska and Peter Pocta}, doi = {10.1016/j.engappai.2025.112286}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112286}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust deep learning based model for denoising phonocardiogram signals in clinical environments}, volume = {161}, year = {2025}, }
• (2025). Transformer-based end-to-end point restoration network for filtering temporarily-static objects from point cloud. EAAI, 161, 112284. (www)

  @article{ , abstract = {The precision of map-based localization systems is pivotal for advanced navigation and autonomous vehicle technologies. This study addresses the challenge of dynamic environments where temporarily-static objects, such as vehicles, introduce occlusion and temporal alterations to the mapping landscape, leading to inaccuracies in mapping and localization. We propose a transformer-based deep learning model that removes temporarily-static objects and refines occluded areas in raw point cloud data, thereby enhancing both mapping and localization. Our method leverages a transformer-based point restoration network trained on a novel dataset that does not require human-driven manual annotation and predicts point clouds free of temporarily-static objects. The approach was validated through experiments demonstrating improved qualitative and quantitative results. The results display the importance of our method by removing temporarily-static objects and refining occluded areas. The proposed method achieves a 26.4 % reduction in Chamfer Distance L1 (CD-L1), 33.1 % reduction in Chamfer Distance L2 (CD-L2), and a 30.2 % relative improvement in the harmonic mean of precision and recall (F-Score) over the baseline network, demonstrating significantly enhanced model for filtering. This research contributes to artificial intelligence (AI)-enabled autonomous navigation by offering a systematic solution to a common problem in dynamic environments, setting a new filtering technology for mapping systems.}, archive = {J_EAAI}, author = {Sabir Hossain and Xianke Lin}, doi = {10.1016/j.engappai.2025.112284}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112284}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based end-to-end point restoration network for filtering temporarily-static objects from point cloud}, volume = {161}, year = {2025}, }
• (2025). Cross-view gait recognition based on discriminative global–local feature representation and learning. EAAI, 161, 112282. (www)

  @article{ , abstract = {Performance of gait recognition can be heavily influenced by deformation of walking silhouettes due to variations of view angle. Cross-view gait recognition is still one of the most challenging problems in pattern recognition community. In this paper, we propose a new cross-view gait recognition algorithm by discriminative global–local feature representation and learning. First, we design cross-view convolutional neural network for learning global gait feature representations underlying binary walking silhouettes sequences. Second, we propose to extract local gait features by partial spatial–temporal sequence and local feature extractors, which can represent the motion characteristics and temporal changes of different human body parts. Both global and local features in our proposed method are extracted from different gait representations to provide spatial–temporal information from diverse aspects, and they are fused on feature level to further reduce the sensitivity to variations of view angle and improve the performance of gait recognition. Finally, we introduce deep metric learning framework and propose a robust, effective, and view-related loss function, named view triplet loss function, to learn discriminative gait features. Compared with triplet loss, the proposed loss function explores hierarchical relationship of the variations of view angles, which achieves better intra-subject compactness and inter-subject separability. Experimental results on CASIA-B, OULP and OUMVLP gait datasets demonstrate the feasibility and advantage of our proposed algorithm.}, archive = {J_EAAI}, author = {Muqing Deng and Yi Zou and Zhi Zeng and Xiaoreng Feng and Yanjiao Wang and Yuan Liu}, doi = {10.1016/j.engappai.2025.112282}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112282}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-view gait recognition based on discriminative global–local feature representation and learning}, volume = {161}, year = {2025}, }
• (2025). A novel clustering-ensemble learning model for day-ahead photovoltaic power forecasting. EAAI, 161, 112281. (www)

  @article{ , abstract = {Accurate day-ahead photovoltaic power forecasting (PPF) is essential for grid scheduling and transaction planning. However, time series prediction models based on historical meteorological and photovoltaic power data often struggle to capture long-term dependencies. Existing research typically employs multivariate regression models to establish mapping relationships between numerical weather prediction (NWP) and photovoltaic power data, yet still exhibits deficiencies in weather pattern recognition and consideration of temporal patterns. Therefore, this paper proposes a novel clustering-ensemble learning model to enhance predictive performance. In the proposed deep time-series clustering algorithm, key features from NWP data are extracted for clustering, addressing the inefficiencies of traditional clustering methods in handling high-dimensional data. A weighted Warp-Euclidean distance is proposed to capture sequence trend homogeneity and spatial similarity, overcoming the limitations of Euclidean distance in time-series data. The ensemble learning model combines the advantages of two gradient boosting tree models to handle nonlinear features and high-dimensional data, while introducing lag features to enhance temporal dynamic learning capability. This approach enhances the stability and mitigates the limitations associated with relying primarily on weather-related features. This study utilizes NWP and measured photovoltaic data from two stations in Hebei Province, China. Results demonstrate superior performance compared to other clustering algorithms and prediction models in day-ahead PPF tasks, achieving R 2 scores of 0.952 and 0.943 on two public photovoltaic datasets, representing improvements of 0.020 and 0.017 over optimal benchmark models. The proposed framework provides theoretical guidance for the stable operation of grid-connected photovoltaic systems.}, archive = {J_EAAI}, author = {Feifei Yang and Xueqian Fu and Zhengshuo Li and Dawei Qiu and Hamed Badihi}, doi = {10.1016/j.engappai.2025.112281}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112281}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel clustering-ensemble learning model for day-ahead photovoltaic power forecasting}, volume = {161}, year = {2025}, }
• (2025). Artificial intelligence-based phonocardiogram signal classification using segment-specific multi-domain features for cardiovascular and arterial disease. EAAI, 161, 112280. (www)

  @article{ , abstract = {Heart diseases are the leading cause of death globally, and early diagnosis coupled with timely treatment can save lives. The human heart produces sounds that reflect the functions of its valves and chambers, with murmurs often indicating abnormal heart behavior. The phonocardiogram (PCG) analysis offers a quick, easy, and affordable diagnostic method for detecting such conditions. The PhysioNet/2016 dataset, containing annotated heart sound recordings, has advanced automated heart sound classification research. Despite the dataset encompassing nine classes of heart diseases, it has primarily been used for binary classification tasks. Multi-class classification is challenging due to overlapping acoustic features among certain heart conditions and signal variability. To address these issues, we implemented segment specific feature extraction across six domains — time, frequency, amplitude, wavelet, spectrum, signal processing — and evaluated various machine learning classifiers. Our findings revealed that the Random Forest classifier outperformed others, leading us to integrate it into our model. On the testing dataset, it attained 94% accuracy, 93% sensitivity and 93% specificity for binary classification Our model achieved 89% accuracy, 80% sensitivity and 98% specificity for nine-class classification including both valve diseases and artery-related conditions.}, archive = {J_EAAI}, author = {Syeda Sana Bukhari and Shahab U. Ansari and Khurram Khan Jadoon and Raja Hashim Ali}, doi = {10.1016/j.engappai.2025.112280}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112280}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-based phonocardiogram signal classification using segment-specific multi-domain features for cardiovascular and arterial disease}, volume = {161}, year = {2025}, }
• (2025). Probabilistic forecasting of non-ferrous metal prices based on outlier treatment algorithms, quantile regression based deep learning and two-phase multi-objective optimization. EAAI, 161, 112277. (www)

  @article{ , abstract = {Accurate forecasting of non-ferrous metal prices plays a crucial role in helping enterprises optimize production, control costs, and manage risks, as well as in enabling countries and industries to formulate resource strategies, ensure economic security, and promote sustainable development. However, existing research still has limitations in outlier detection, uncertainty analysis, and model optimization, making it difficult to meet the forecasting needs in complex market environments. To fill these gaps, this study proposes a novel combined probabilistic forecasting framework for non-ferrous metal prices. First, this study utilizes outlier treatment algorithms to achieve robust outlier detection and correction. Second, multiple quantile regression-based deep learning models are established for probabilistic forecasting. Subsequently, a two-phase multi-objective optimization strategy is proposed. Specifically, the first phase is designed to optimize the upper quantile half-interval and lower quantile half-interval of each quantile regression-based deep learning model, while the second phase is used to optimize the combined weights of the aforementioned single model to synergistically enhance interval reliability and prediction accuracy. Finally, silver and aluminum futures prices are used as illustrative case studies. The experimental results show that the proposed model outperforms the benchmark method in terms of interval coverage and prediction accuracy. This study provides a scientific basis and reference for industrial production optimization, investment decision-making and resource management planning.}, archive = {J_EAAI}, author = {Pei Du and Fangrui Gui and Lijie Shan and Qianyi Xing and Jianzhou Wang}, doi = {10.1016/j.engappai.2025.112277}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112277}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Probabilistic forecasting of non-ferrous metal prices based on outlier treatment algorithms, quantile regression based deep learning and two-phase multi-objective optimization}, volume = {161}, year = {2025}, }
• (2025). Identifying nonlinear roll damping and restoring parameters via physics-informed neural network. EAAI, 161, 112275. (www)

  @article{ , abstract = {To effectively characterize a ship’s roll motion, developing a robust dynamic model that incorporates accurate damping and restoring parameters is essential. Given that only limited free-decay measurements have been measured, an efficient physics-informed neural network-based approach has been developed to simultaneously determine the nonlinear damping and restoring parameters associated with ship rolling. Rather than relying on a large dataset to train a neural network for parameter identification, the proposed method incorporates the physical equations of roll motion into the residual networks, ensuring that the identified parameters adhere to their physical interpretation. Three numerical cases, encompassing computational fluid dynamics (CFD) analyses, and laboratory experiments, are presented to show the superior performance of the developed method. Key findings indicate that the proposed method effectively identifies the investigated parameters with high precision while maintaining consistent performance across various initial roll angle conditions. Compared to traditional methods, the proposed approach offers significant advantages, including reduced reliance on large datasets, automated parameter identification without manual tuning, and the ability to incorporate physical constraints directly into the learning process. These features make the method particularly suitable for real-world applications where data is sparse or costly to obtain.}, archive = {J_EAAI}, author = {Shuai Cong and Jinwei Sun and Qianying Cao and Changhong Zhi and Shixuan Liu}, doi = {10.1016/j.engappai.2025.112275}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112275}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identifying nonlinear roll damping and restoring parameters via physics-informed neural network}, volume = {161}, year = {2025}, }
• (2025). Anomaly detection in offshore open radio access network using long short-term memory models on a novel artificial intelligence-driven cloud-native data platform. EAAI, 161, 112274. (www)

  @article{ , abstract = {The Radio Access Network (RAN) is a critical component of modern telecommunications infrastructure, currently evolving towards disaggregated and open architectures. These advancements are pivotal for integrating intelligent, data-driven applications aimed at enhancing network reliability and operational autonomy through the introduction of cognitive capabilities, as exemplified by the emerging Open Radio Access Network (O-RAN) standards. Despite its potential, the nascent nature of O-RAN technology presents challenges, primarily due to the absence of mature operational standards. This complicates the management of data and intelligent applications, particularly when integrating with traditional network management and operational support systems. Divergent vendor-specific design approaches further hinder migration and limit solution reusability. These challenges are compounded by a skills gap in telecommunications business-oriented engineering, which remains a key barrier to effective O-RAN deployment and intelligent application development. To address these challenges, Boldyn Networks developed a novel cloud-native data analytics platform, specifically designed to support scalable Artificial Intelligence (AI) integration within O-RAN deployments. This platform underwent rigorous testing in real-world scenarios, and applied advanced AI techniques to improve operational efficiency and customer experience. Implementation involved adopting Development Operations (DevOps) practices, leveraging data lakehouse architectures tailored for AI applications, and employing sophisticated data engineering strategies. The platform successfully addresses connectivity challenges inherent in real-world offshore wind farm deployments using Long Short-Term Memory (LSTM) models for anomaly detection in network connectivity. After integrating the LSTM models into the network control, more than 90 percent of connectivity issues were reduced in runtime. This marks a step toward autonomous, self-organizing, and self-healing networks.}, archive = {J_EAAI}, author = {Abdelrahim Ahmad and Peizheng Li and Robert Piechocki and Rui Inacio}, doi = {10.1016/j.engappai.2025.112274}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112274}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in offshore open radio access network using long short-term memory models on a novel artificial intelligence-driven cloud-native data platform}, volume = {161}, year = {2025}, }
• (2025). Structural design through reinforcement learning. EAAI, 161, 112273. (www)

  @article{ , abstract = {This paper introduces the Structural Optimization gym (SOgym), a novel open-source Reinforcement Learning (RL) environment designed to advance machine learning in Topology Optimization (TO). SOgym enables RL agents to generate physically viable and structurally robust designs by integrating the physics of TO into the reward function. To enhance scalability, SOgym leverages feature-mapping methods as a mesh-independent interface between the environment and the agent, allowing efficient interaction with the design variables regardless of mesh resolution. Baseline results use a model-free Proximal Policy Optimization agent and a model-based DreamerV3 agent. Three observation space configurations were tested. The Topology Optimization game ( TopOpt game) inspired configuration, an interactive educational tool for designing structures to minimize compliance under volume constraints, performed best in terms of performance and sample efficiency. The 100 million parameter DreamerV3 model (DreamerV3-100M) produced structures with compliance values approximately 54 % higher than those from traditional optimization methods and a 0 % disconnection rate, an improvement over supervised learning approaches that often struggle with disconnected load paths. When comparing the learning rates of the agents to those of engineering students from the TopOpt game experiment, the DreamerV3-100M model shows a learning rate approximately four orders of magnitude lower, an impressive feat for a policy trained from scratch through trial and error. These results suggest RL's potential to solve continuous TO problems and its capacity to explore and learn from diverse design solutions. SOgym provides a platform for developing RL agents for structural design challenges and is publicly available to support research in the field.}, archive = {J_EAAI}, author = {Thomas Rochefort-Beaudoin and Aurelian Vadean and Niels Aage and Sofiane Achiche}, doi = {10.1016/j.engappai.2025.112273}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112273}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Structural design through reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Single-channel selection for epileptic seizure identification through a custom machine learning model. EAAI, 161, 112272. (www)

  @article{ , abstract = {Epilepsy affects millions of people worldwide, triggering undesirable motor and sensory events that drastically impair the quality of life of those affected. Detecting this condition through a monitoring system can be extremely useful in identifying epileptic seizures without relying exclusively on a specialist. Our goal is to identify these seizures using personalized supervised machine learning models – eXtreme Gradient Boosting – for each patient, employing only one electroencephalogram input channel. Our methodology explores the relationship between the selected input channel and the topographic location of epileptic seizures modeled for the patient. The results revealed notable accuracy performances for the three patients investigated: 100%, 99%, and 88%, after applying the proposed time filter that checks for the presence/absence of a seizure every 3 s. Furthermore, it was possible to observe that the results are consistent with the affected area in each patient, demonstrating the effectiveness of the method in selecting the single channel. This approach demonstrates the feasibility of detecting convulsive seizures using only one input channel and underscores the need to extend the methodology to more patients.}, archive = {J_EAAI}, author = {Jusciaane Chacon Vieira and Ignacio Sanchez-Gendriz and Luiz Affonso Guedes}, doi = {10.1016/j.engappai.2025.112272}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112272}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Single-channel selection for epileptic seizure identification through a custom machine learning model}, volume = {161}, year = {2025}, }
• (2025). IMobileTransformer: A fusion-based lightweight model for rice disease identification. EAAI, 161, 112271. (www)

  @article{ , abstract = {Rice blast, sheath blight, leaf scald, bacterial leaf blight, and brown spot severely threaten rice yield. To address the limitations of current deep learning methods in rice disease recognition, particularly their insufficient integration of local and global features, this study proposes an Improved MobileTransformer (IMobileTransformer) model. The proposed architecture synergistically combines MobileNet’s strengths in local feature extraction and lightweight architecture with Transformer’s superior capability in global information processing. Specifically, the model is designed with three functional branches: a) a MobileNet branch utilizing inverted residual structure with depthwise separable convolution layers to reduce parameters and computational complexity, b) a Transformer branch modified from Swin-Transformer architecture, where the Multilayer Perceptron (MLP) layer is enhanced by splitting input feature channels through an Inception-based structure to maintain global feature extraction efficiency while minimizing computational overhead, and c) a feature fusion branch that concatenates reshaped outputs from both branches through channel-wise stacking, enabling effective integration of local and global representations. Experimental results show that compared to classical models such as MobileNetV3-Large, EfficientNet-B0, Vision Transformer Base/16 (ViT-B/16), Shifted Window Transformer (Swin-Transformer), Tiny Vision Transformer (TinyViT), Mobile Vision Transformer (MobileViT), LocalViT-S, IMobileTransformer achieves a recognition accuracy of 99.62% for rice diseases, with improvements of 1.71%, 0.91%, 38.09%, 4.17%, 1.99%, 1.5% and 0.42%, respectively, providing an effective solution for rice disease recognition.}, archive = {J_EAAI}, author = {Yang Lu and Haoyang Zhou and Peng Wang and Erzhi Wang and Gongfa Li and Tongjian Yu}, doi = {10.1016/j.engappai.2025.112271}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112271}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {IMobileTransformer: A fusion-based lightweight model for rice disease identification}, volume = {161}, year = {2025}, }
• (2025). Personalized and safe medication recommendation based on convolutional neural network and transformer architecture. EAAI, 161, 112267. (www)

  @article{ , abstract = {With the accumulation of electronic health records (EHRs), artificial intelligence (AI) based medical services such as medication recommendation (MR) has aroused widespread concern. However, existing drug recommendation models suffer from inadequate patient representation and adverse drug–drug interactions (DDIs). To address these challenges, we propose an AI-based personalized and safe medication recommendation method based on convolutional neural network and transformer architecture (CT-PASMR). Specifically, convolutional neural network (CNN) and transformer are combined in parallel (CAT) to model local relationships in a patient’s single visit and long-term dependencies in sequential EHR data, respectively. Subsequently, graph attention networks (GATs) are deployed to capture drug co-occurrences and adverse DDIs with various weights, generating safe drug representations. Moreover, a joint loss function is introduced to balance accuracy and safety in CT-PASMR. Finally, the experimental results on MIMIC (Medical Information Mart for Intensive Care)-III and MIMIC-IV datasets demonstrate that CT-PASMR achieves competitive performance on seven evaluation metrics such as DDI rate, Jaccard index and F1 score, compared with nine state-of-the-art (SOTA) baseline models. Ultimately, ablation studies and further analysis confirm the efficacy of CATs and GATs in providing personalized and safe medication recommendations. Code, qualitative results, and trained weights will be available at the link: https://github.com/gefengru/CT-PASMR .}, archive = {J_EAAI}, author = {Fengru Ge and Xiaomei Yu and Xue Li and Xingxu Fan and Yanjie Zhao}, doi = {10.1016/j.engappai.2025.112267}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112267}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalized and safe medication recommendation based on convolutional neural network and transformer architecture}, volume = {161}, year = {2025}, }
• (2025). Contrastive learning with similarity enhancement for dimensionality reduction. EAAI, 161, 112266. (www)

  @article{ , abstract = {Dimensionality reduction aims to reduce the number of dimensions while preserving crucial information. As a self-supervised learning approach, contrastive learning provides a novel perspective for dimensionality reduction. However, most contrastive learning methods focus on optimizing similarity, which have limitations in reducing feature redundancy. Moreover, the dependence on negative pairs introduces computational overhead. To address these problems, we propose Contrastive Learning with Similarity Enhancement for Dimensionality Reduction (CLSDR), which integrates neighborhood embedding into the contrastive learning framework. Specifically, CLSDR uses k -nearest neighbors sampling to construct positive pairs. We design a multi-level loss function that captures the diversity of data while maintaining the consistency of local and global features. In addition, we propose a nonlinear similarity optimization mechanism based on logarithmic smoothing, which adjusts the gradient of similarity loss, improving the stability during the training process. Experimental results demonstrate that CLSDR significantly outperforms several state-of-the-art methods. Especially, on the Street View House Numbers dataset, CLSDR achieves 66.7% and 62.3% Top-1 accuracy with two classifiers in 64-dimensional embedding space, which have 10.6% and 38.4% improvement over the best competing approach. Thus, CLSDR exhibits strong robustness and scalability across different dimensions.}, archive = {J_EAAI}, author = {Qi Yang and Changpeng Wang and Linlin Feng and Lizhen Ji and Jiangshe Zhang}, doi = {10.1016/j.engappai.2025.112266}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112266}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrastive learning with similarity enhancement for dimensionality reduction}, volume = {161}, year = {2025}, }
• (2025). A multi-modal approach for detecting drivers’ distraction using bio-signal and vision sensor fusion in driver monitoring systems. EAAI, 161, 112265. (www)

  @article{ , abstract = {According to a report by the World Health Organization (WHO), approximately 1.3 million people lose their lives annually owing to traffic accidents. The majority of road traffic accidents stem from driver negligence. Recently, there has been a growing interest in utilizing deep learning and machine learning technologies to enhance the safety and efficiency of road traffic, with the aim of addressing issues arising from driver inattentiveness. Most studies focus on detecting abnormal driver behavior using driving sensors or driver images; however, they often overlook physiological factors such as the driver's bio-signals. Considering that the driver's state, including fatigue, stress, and concentration, can significantly affect driving safety, it is crucial to build models that consider biometric information. Therefore, this study proposes a multi-modal transformer model called Bio-Vision Transformer (BiViT) that comprehensively considers both driver bio-signals and images. The BiViT model uses a vision transformer to extract features from driver images and employs a time-series transformer to capture features from the driver's bio-signals. In addition, the interactions between the extracted features are modeled, and the joint fusion method is employed as the feature-fusion approach. To validate the proposed model, performance comparisons and analyses were conducted using commonly used models in image analysis. The experimental results demonstrated that the proposed BiViT model exhibited high performance, with an accuracy of 0.91 and a harmonic mean of precision and recall (F1-score) of 0.91, surpassing the performance of the comparison models.}, archive = {J_EAAI}, author = {Byeongjoon Noh and Myeongseok Park and Yechan Han and Jaeyun Kim}, doi = {10.1016/j.engappai.2025.112265}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112265}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-modal approach for detecting drivers’ distraction using bio-signal and vision sensor fusion in driver monitoring systems}, volume = {161}, year = {2025}, }
• (2025). Graph neural network with generative adversarial training for node classification on class imbalanced data. EAAI, 161, 112264. (www)

  @article{ , abstract = {Node classification in class-imbalanced graph data remains a critical challenge, as traditional graph neural networks (GNNs) either assume class-balanced graph structures or inadequately address class imbalance. This often results in predictive bias, where majority classes are favored while minority classes are underrepresented. To overcome this limitation, this study introduces a novel graph neural network with generative adversarial training (GNN-GAN), where the GNN extracts latent features from input node attributes balanced through data synthesis using a conditional generative adversarial network (GAN) and data fusion strategy. The GNN and GAN are trained synchronously to ensure GAN synthesizes samples that match real data distribution, while GNN adjusts to the quality of synthesized data in a timely manner. A data fusion strategy combines synthetic and real samples to mitigate class imbalance and maintain classification accuracy. Experiments on several benchmark graph datasets demonstrate that the GNN-GAN consistently outperforms state-of-the-art baselines. A comprehensive ablative study further validates the advantages of the synchronized training procedure, offering insights into the model's robustness across graph datasets with varying structures and imbalance ratios.}, archive = {J_EAAI}, author = {Xiaoqi Zhou and Peixin Shi}, doi = {10.1016/j.engappai.2025.112264}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112264}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural network with generative adversarial training for node classification on class imbalanced data}, volume = {161}, year = {2025}, }
• (2025). Diffusion-based feature absorption approach for improving lightweight object detectors in adverse weather conditions. EAAI, 161, 112263. (www)

  @article{ , abstract = {Although cutting-edge lightweight detectors have achieved desirable performance in favorable weather conditions, they again failed to accurately identify objects within low-quality scenes captured in inclement environments, especially during rainy nighttime. These detectors face difficulties learning beneficial information for detection because objects are obscured by rain and low-light conditions. To address the aforementioned challenges, we introduce an innovative and effective Diffusion-based Feature Absorption Learning (Diff-FAL) approach, to reinforce the performance of lightweight detection models in nighttime with rain interference. By developing an unsupervised training strategy based on the diffusion model, the proposed approach assists lightweight detectors absorb useful features from degraded images. To this end, our Diff-FAL framework comprises three distinct subnetworks: a Feature Optimization (FO) subnetwork, a Feature Mutation (FM) subnetwork, and a Lightweight Detection (LD) subnetwork. The FO subnetwork is designed to produce sharp, detailed features from rainy night images and the FM subnetwork is developed to transfer those features to the LD subnetwork. Comprehensive experiments on various datasets confirmed the superiority of our model, outperforming the compared method by up to 28.01% and 30.16% in accuracy on two published datasets: the rainy nighttime (RNT) and real rainy images (rRain) datasets, respectively, while maintaining high-speed performance during inference.}, archive = {J_EAAI}, author = {Quoc-Viet Hoang and Trung-Hieu Le}, doi = {10.1016/j.engappai.2025.112263}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112263}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diffusion-based feature absorption approach for improving lightweight object detectors in adverse weather conditions}, volume = {161}, year = {2025}, }
• (2025). A multi-modal model for removal of crack image shadow based on language prior information. EAAI, 161, 112262. (www)

  @article{ , abstract = {Computer-vision based crack detection highly depends on the quality of images and it remains to be a challenging task due to the effects of shadow. The utilization of deep learning algorithm, guided by prior information, has been demonstrated effective in the removal of shadows. However, it has limitations in handling images with shadowed cracks. A deep learning method incorporating language prior information and a multi-modal model is proposed in this study to improve the removal of shadow effects in restoring the crack images. Natural language processing (NLP) is used to extract the language prior information from image text description to enhance the crack recognition, and OTSU method is used to obtain the image prior information. A novel multi-modal model, named as the Text-based Shadow Removal Network (TSRNet), for the removal of shadow in crack image, is proposed to have better crack restoration capability. Bayesian optimization approach is also employed to optimize the network architecture improving the prediction precision of TSRNet. The proposed framework for shadow removal is verified using an open-source shadowed concrete crack image dataset and a new dataset from experiment. Results indicate that the language prior information can enhance the TSRNet model with better performances of shadow removal for real-world crack images compared with other existing models.}, archive = {J_EAAI}, author = {Gang Liu and Xuming Li and Qingshan Yang and S.S. Law and Changjun Deng}, doi = {10.1016/j.engappai.2025.112262}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112262}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-modal model for removal of crack image shadow based on language prior information}, volume = {161}, year = {2025}, }
• (2025). Development and implementation of a hybrid visual prediction algorithm for robotic smart tomato harvesting. EAAI, 161, 112261. (www)

  @article{ , abstract = {This paper introduces new algorithms designed specifically for smart tomato harvesting, which combines predictions from two independently trained YOLOv8 (You Only Look Once version 8) models, each specialized on different datasets to detect tomatoes and classify their ripeness stages—ripe, unripe, and semi-ripe—while also computing their center points. To enhance detection accuracy under varying field conditions, multiple fusion strategies were developed, including a hybrid algorithm that integrates union and confidence-weighted summation methods. The Hybrid Algorithm achieved the best performance, surpassing the original models and other fusion techniques. It attained F1 scores of 0.697 and 0.694 at confidence thresholds of 0.5 and 0.9, respectively, compared to the original models' F1 scores of 0.670 and 0.648 at confidence 0.5, and 0.481 and 0.497 at confidence 0.9. This corresponds to an improvement of 4.0 % and 7.6 % at confidence 0.5, and 44.3 % and 39.6 % at confidence 0.9, demonstrating the hybrid algorithm's stability and superiority, particularly at higher thresholds. Furthermore, the system utilizes a high-resolution RGB (Red, Green, Blue) camera for real-time image capture, enhancing model performance in complex agricultural environments. This study validates the effectiveness of confidence-based fusion in developing robust and accurate vision systems for precision agriculture.}, archive = {J_EAAI}, author = {Giuseppe Carbone and Angad Singh Gurtatta and Dmitry Malyshev}, doi = {10.1016/j.engappai.2025.112261}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112261}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development and implementation of a hybrid visual prediction algorithm for robotic smart tomato harvesting}, volume = {161}, year = {2025}, }
• (2025). Multi-modal multi-level feature representation learning for flow pattern identification of oil-water two-phase flow. EAAI, 161, 112260. (www)

  @article{ , abstract = {To tackle the difficulty of the traditional experimental methods in real-time monitoring and identification of the flow process, this paper introduces a novel flow pattern identification method of the vertical oil-water two-phase flow based on multi-modal multi-level feature representation. The one-dimensional electromagnetic signals are encoded into two-dimensional feature spaces to explore their structural complexity, evolutionary probability laws and nonlinear characteristics in multimodal domain, thereby generating a multi-modal representation of the electromagnetic signals. Subsequently, a multi-modal multi-level feature fusion network is developed for flow pattern identification network, which flexibly leverages effective information across different modalities and levels, thereby enhancing the identifying accuracy. Experimental results demonstrate that the proposed method achieves high accuracy on the constructed multi-modal dataset, proving its feasibility and effectiveness in identifying the flow pattern of the oil-water two-phase flow in vertical pipes.}, archive = {J_EAAI}, author = {Weihang Kong and Yaohan Chi and He Liu and Hongbao Tang and He Li}, doi = {10.1016/j.engappai.2025.112260}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112260}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal multi-level feature representation learning for flow pattern identification of oil-water two-phase flow}, volume = {161}, year = {2025}, }
• (2025). Defining and evaluating decision and composite risk in language models applied to natural language inference. EAAI, 161, 112253. (www)

  @article{ , abstract = {Despite their impressive performance, large language models (LLMs) are known to pose important risks. One such set of risks arises from misplaced confidence, whether over-confidence or under-confidence, that the models have in their inference. While the former is well studied, the latter is not, leading to an asymmetry in understanding the comprehensive risk of the model based on misplaced confidence. In this paper, we address this asymmetry by defining two types of risk (decision and composite risk), and proposing an experimental framework consisting of a two-level inference architecture and appropriate metrics for measuring such risks in both discriminative and generative LLMs. The first level relies on a decision rule that determines whether the underlying language model should abstain from inference. The second level (which applies if the model does not abstain) is the model’s inference. This framework has direct implications for error-sensitive LLM-based engineering applications where reliable decision-making is critical, such as healthcare and finance. Through detailed experiments on four natural language commonsense reasoning datasets using both an open-source ensemble-based transformer model and a generative LLM, we demonstrate the practical utility of our evaluation framework. Our results show that the framework can get an LLM to confidently respond to an extra 20.1% of low-risk inference tasks that other methods might misclassify as high-risk, and skip 19.8% of high-risk tasks, which would have been answered incorrectly.}, archive = {J_EAAI}, author = {Ke Shen and Mayank Kejriwal}, doi = {10.1016/j.engappai.2025.112253}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112253}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Defining and evaluating decision and composite risk in language models applied to natural language inference}, volume = {161}, year = {2025}, }
• (2025). Advanced machine learning techniques for predicting wear performance in graphene oxide particulate interpenetrating polymer network composites. EAAI, 161, 112252. (www)

  @article{ , abstract = {This research investigates the wear behavior of hybrid polymeric composites made from synthetic glass and natural cotton fibers, reinforced with varying proportions of Graphene Oxide (GO) (0 %, 1 %, 3 %, 5 %, 7 %, 9 %). The effect of fiber arrangement and Graphene Oxide (GO) incorporation on wear rate and Coefficient of Friction (CoF) was evaluated using the Pin-On-Disk method, with analysis based on Taguchi's L 32 Orthogonal Array. The optimal parameters were found at 6 min, 5 % GO, 300 revolutions per minute (rpm) speed, 20 mm (mm) track diameter, and 10 N (N) load, achieving a minimum wear rate of 0.612 × 10 −4 cubic millimeters per newton-meter (mm 3 /N-m) and a CoF of 0.151. Predictive modeling was performed to predict the wear rate and coefficient of friction using supervised machine learning algorithms, including Linear Regression, Decision Tree, and Random Forest, to forecast material behavior. Performance evaluation using Confusion Matrix, Distribution Analysis, and various metrics showed that the Decision Tree model excelled, achieving near-perfect predictive power with a Mean Squared Error (MSE) of 0 and an R-squared value of 0.9999. The model demonstrated 100 % accuracy, with precision, recall, and F1-scores all equal to 1. This research demonstrates the effectiveness of combining natural and synthetic fibers with GO, along with the predictive power of machine learning in optimizing material properties.}, archive = {J_EAAI}, author = {Eastus Russel and S. Madhu and Judy S and Edwin Geo Varuvel and G.B. Santhi and G. Suresh and Femilda Josephin J.S and Mohammed F. Albeshr and Farzad Kiani}, doi = {10.1016/j.engappai.2025.112252}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112252}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced machine learning techniques for predicting wear performance in graphene oxide particulate interpenetrating polymer network composites}, volume = {161}, year = {2025}, }
• (2025). Health state evaluation and analysis of equipment considering multi-scale data fusion. EAAI, 161, 112251. (www)

  @article{ , abstract = {Current researches of health state evaluation of equipment have increasingly emphasized the integration of data and knowledge to improve evaluation accuracy and interpretability, which leads to the emergence of hybrid information-based methods as a research hotspot. However, there are issues with redundant health indicators, multi-scale data with different features and measurement units, and interpretability measures in the evaluation. In this paper, an interpretable health state evaluation and analysis method of equipment is proposed based on trustworthy evidential reasoning rule (TERR). To deal with redundancy in health indicators, a multi-scale data analysis model is proposed based on the clustering analysis and Kruskal-Wallis test. It can select the optimal health indicators that effectively preserve the original evaluation information while reducing the model complexity. To integrate the selected indicators with uncertainty, a TERR-based multi-scale data fusion model is proposed, where the evidence weight, reliability, and trustworthiness are simultaneously considered. Also, an interpretable parameter optimization model is constructed to alleviate the uncertainty in initial evidential parameters. To study the interpretability of TERR, two sensitivity analysis methods are proposed, and the performance coefficient matrix of the model output to the perturbation is derived mathematically. This shows how the interpretability of TERR is enhanced and which external parameter has the greatest impact on outputs. Finally, a case study of health state evaluation of laser inertial measurement unit (LIMU) validates the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Shuai-Wen Tang and Jiang Jiang and Jian-Bin Sun and Zhuo-Ting Yu}, doi = {10.1016/j.engappai.2025.112251}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112251}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health state evaluation and analysis of equipment considering multi-scale data fusion}, volume = {161}, year = {2025}, }
• (2025). Unmanned surface vehicle autonomous racing and obstacle avoidance with robust adversarial deep reinforcement learning. EAAI, 161, 112250. (www)

  @article{ , abstract = {This paper presents an autonomous racing control method for Unmanned Surface Vehicles (USVs) based on robust adversarial deep reinforcement learning (ADRL) algorithm, which leverages the strengths of both deep reinforcement learning and adversarial training. Adversarial obstacles and various tracks are employed to train a policy, so the proposed method can enhance the robustness and generalization of autonomous USV racing while ensuring effective obstacle avoidance. A simulation environment for USV racing was developed to conduct the experiments with Unity3D. The performance of the proposed method in handling diverse track scenarios and obstacle avoidance is demonstrated through simulations. Quantitative results show that ADRL achieves dramatic improvements over baseline DRL methods: collision rates are reduced by 99.85%, task completion rates are improved by 73.75%, and navigation time efficiency is enhanced by approximately 3% while maintaining superior safety performance.}, archive = {J_EAAI}, author = {Jingyu Liu and Yuanda Wang and Changyin Sun}, doi = {10.1016/j.engappai.2025.112250}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112250}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unmanned surface vehicle autonomous racing and obstacle avoidance with robust adversarial deep reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Multiscale constitutive modeling of anisotropic plasticity: Coupling the visco-plastic self-consistent model with the recurrent neural network and its implementation in finite element analysis. EAAI, 161, 112249. (www)

  @article{ , abstract = {The anisotropic and nonlinear strain-path-dependent nature of metal plasticity poses a major challenge for accurate constitutive modeling in finite element (FE) analysis. Traditional macroscale models are easily implemented but lack accuracy, while crystal plasticity (CP) models offer high fidelity at the cost of computational efficiency. To bridge this gap, we propose a deep neural network smart constitutive (DNNSC) framework that combines the visco-plastic self-consistent (VPSC) model with a gated recurrent unit (GRU) network. A VPSC model calibrated on pure aluminum generated 14,000 strain-paths for training GRU-based network. The optimized model has a prediction accuracy of up to 96 % on unknown strain-paths. Subsequently, the DNNSC model was implemented into the FE analysis through Fortran programming, and a benchmark simulation for thin sheet stamping was successfully performed. The simulation results demonstrated that the DNNSC model significantly improved prediction performance compared to conventional macroscale constitutive models. Especially, the ear height and plate thickness were accurately predicted with an accuracy of 91.85 % and 95.84 %, compared to only 68.85 % and 86.59 % achieved by the Yld model. Meanwhile, the simulation time was reduced to approximately one-tenth that of the fully coupled CP model, because the latter required calculating and homogenizing the mechanical responses of hundreds of grains at each integration point during the simulation. The DNNSC framework bridges the gap between CP models and FE simulations of plastic forming and breaks down the barrier between modeling and practical application. Furthermore, this framework can be extended to other materials by re-calibrating VPSC parameters and fine-tuning DNN parameters.}, archive = {J_EAAI}, author = {Ziwei Zhou and Liang Cheng and Huaidong Song and Haijing Guo and Ruolin Li and Lingyan Sun and Bin Tang}, doi = {10.1016/j.engappai.2025.112249}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112249}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiscale constitutive modeling of anisotropic plasticity: Coupling the visco-plastic self-consistent model with the recurrent neural network and its implementation in finite element analysis}, volume = {161}, year = {2025}, }
• (2025). An online reinforcement learning method to improve control adaptability in robot-aided rehabilitation. EAAI, 161, 112248. (www)

  @article{ , abstract = {Rehabilitation robotics enables consistent and personalized therapy but still relies on complex, expert-driven tuning of control parameters. To address this, a reinforcement learning strategy based on Q-learning is proposed to autonomously adapt key parameters during upper-limb rehabilitation, without requiring prior task-specific knowledge. A systematic evaluation is conducted across combinations of control parameters (radial stiffness and execution time), performance-based reward functions (pointing accuracy and movement smoothness), and exploration strategies ( ɛ - greedy and Upper Confidence Bound). The Q-learning agent selects discrete actions (increase, decrease, or maintain the current value) for each control parameter, enabling real-time adaptation based on observed performance. The method is validated using a Kuka robotic arm in experiments involving 16 right-handed healthy subjects (13 males, 3 females) and 8 right-handed individuals simulating impaired motor behavior (5 males, 3 females). Motion signals are acquired through internal robot sensors, while a wearable physiological monitoring system define the Q-learning agent state. Reward improvement and exploration ratio are analyzed as key performance indicators and statistically compared across all tested conditions using the Mann–Whitney test. The results demonstrate that the proposed algorithm effectively adjusts control parameters online, with performance influenced by the reward function, exploration strategy, and selected control actions. Reward improvements of 0 . 11 ± 0 . 09 ( ɛ - greedy , reward based on pointing ability) and 0 . 13 ± 0 . 11 (reward based on smoothness, Upper Confidence Bound strategy) were observed in healthy subjects, indicating enhancements in pointing accuracy and movement smoothness. In simulated pathological cases, improvements of 0 . 08 ± 0 . 13 and 0 . 06 ± 0 . 16 were observed, respectively.}, archive = {J_EAAI}, author = {Rita Molle and Christian Tamantini and Clemente Lauretti and Emilio Maria Romano and Loredana Zollo}, doi = {10.1016/j.engappai.2025.112248}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112248}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An online reinforcement learning method to improve control adaptability in robot-aided rehabilitation}, volume = {161}, year = {2025}, }
• (2025). Fast prediction and compensation of curing deformation behaviours of composite parts with complex geometry based on neural operator on riemannian manifolds. EAAI, 161, 112247. (www)

  @article{ , abstract = {Controlling the curing deformation of composite parts is becoming increasingly challenging with the ever-increasing performance requirements of aerospace equipment. Mould surface compensation, which adjusts the mould surface to minimise the discrepancy between the cured part geometry and the nominal part geometry, has become a primary deformation control way in engineering. The existing mirror compensation methods focus on the deformation dominated by spring-in and are challenging for complex deformation modes. Surrogate model-based shape optimisation provides a feasible idea, but establishing a surrogate model to predict curing deformation fields on complex part geometries remains a challenge. Therefore, this study explores a novel neural operator-driven framework for fast curing deformation prediction and compensation. A clustering-based deformation field segmentation method is proposed to manipulate the mould surface morphing using limited design variables. The neural operator on Riemannian manifolds is introduced for the first time to establish the surrogate model between the mould surface and the curing deformation fields on complex part geometries. To control the global error distribution of the composite part, two error metrics are designed to optimise the mould surface by genetic algorithm. The verification results show that the proposed framework exhibits significant potential in predicting and compensating for the curing deformation field of composite parts with complex geometry.}, archive = {J_EAAI}, author = {Lu Chen and Yingguang Li and Jingyan Su and Weiwei Xu and Lin Hu and Gengxiang Chen and Xu Liu and Xiaozhong Hao}, doi = {10.1016/j.engappai.2025.112247}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112247}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast prediction and compensation of curing deformation behaviours of composite parts with complex geometry based on neural operator on riemannian manifolds}, volume = {161}, year = {2025}, }
• (2025). Enhanced detection of acute leukemia: A hybrid machine learning framework with adaptive weight-optimized level set evolution. EAAI, 161, 112244. (www)

  @article{ , abstract = {In this research, a novel, multistage framework is developed for the automated detection of Acute Lymphoblastic Leukemia (ALL) and Acute Myeloid Leukemia (AML), addressing challenges due to the inclusion of overlapping cells, noise, and unwanted cells. It integrates an Adaptive Weight-Optimized Level Set Evolution (AWOLSE) scheme to ensures precise and accurate segmentation, coupled with a marker-controlled watershed algorithm to improve performance in regions of cell overlap and contact. It ensures more reliable cell boundary delineation. For classification, a hybrid model combining the strengths of Random Forest (RF) and Support Vector Machine (SVM) is employed that delivering superior performance by leveraging the complementary advantages of these classifiers. Furthermore, feature extraction with the Gray Level Co-occurrence Matrix (GLCM), followed by feature selection with Principal Component Analysis (PCA), aids in the identification of relevant features. The suggested methodology beats its competitors, providing the best ALL identification results on the Acute Lymphoblastic Leukemia Image Database (ALLIDB), with 99.07% accuracy, 97.96% recall, 100.00% specificity, and 100.00% precision. Similarly, on the American Society of Hematology (ASH) database, the technique achieves superior AML identification results with 96.25% accuracy, 95.00% recall, 97.50% specificity, and 97.44% precision.}, archive = {J_EAAI}, author = {Pradeep Kumar Das and Adyasha Sahu and Sukadev Meher and Rutuparna Panda and Ajith Abraham}, doi = {10.1016/j.engappai.2025.112244}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112244}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced detection of acute leukemia: A hybrid machine learning framework with adaptive weight-optimized level set evolution}, volume = {161}, year = {2025}, }
• (2025). Reinforcement learning-based secure tracking control for nonlinear interconnected systems: An event-triggered solution approach. EAAI, 161, 112243. (www)

  @article{ , abstract = {This paper proposes a secure tracking control method for nonlinear interconnected systems based on reinforcement learning, addressing both security constraints and mismatched conditions in such systems. By utilizing system augmentation techniques, the tracking control problem is reformulated into a stabilization problem for the augmented system, simplifying the original control task. Drawing inspiration from the Chinese philosophical principle, we further transform the problem into a zero-sum game framework. Moreover, a control barrier function (CBF) is incorporated into a cost function to ensure that system trajectories remain within a predefined safe region. An event-triggered mechanism is introduced, and an event-based safety Hamilton–Jacobi–Isaacs (HJI) equation is established. An adaptive single evaluation network is designed, leveraging experience replay techniques to solve the HJI equation. Finally, the Lyapunov method is employed to prove the uniform ultimate boundedness of both the tracking error and the weight error. The effectiveness of the proposed method is validated through numerical examples.}, archive = {J_EAAI}, author = {Chunbin Qin and Suyang Hou and Mingyu Pang and Zhongwei Wang and Dehua Zhang}, doi = {10.1016/j.engappai.2025.112243}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112243}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning-based secure tracking control for nonlinear interconnected systems: An event-triggered solution approach}, volume = {161}, year = {2025}, }
• (2025). A quantized subtraction-convolution network for industrial lightweight edge interpretable diagnosis. EAAI, 161, 112241. (www)

  @article{ , abstract = {In addressing the prevalent challenges of delayed fault diagnosis, deployment complexity of advanced deep learning models on low-cost edge devices, and limited model interpretability, this study proposed an edge-cloud collaborative interpretable diagnosis based on quantized subtraction-convolution network. Firstly, an interpretable quantized subtraction-convolution network with lightweight three-layer structure is designed. Inspired by the adaptive spectral subtraction, a learnable sparse pulse kernel is designed to extract the fault feature as the subtraction layer. Subsequently, the convolution and classification layers are then integrated to produce interpretable results for efficient identification. To facilitate deployment and updates on edge devices, the quantized subtraction-convolution network is decomposed into a lightweight edge architecture and corresponding parameters. It can be deployed on edge devices, and an edge-cloud collaborative framework addresses its training and compression. Considering the sparse characteristics of quantized subtraction-convolution network, a sparse pulse quantization strategy and quantization-aware training technique were developed to compress the model parameters. Finally, a low-cost edge fault diagnosis node prototype with quantized subtraction-convolution network is designed for real-time edge fault diagnosis. Experiments shown that the proposed method achieved average accuracy of 99.88 percent with compression ratio of 9.5. The memory usage, floating-point operations per second, and average power consumption are respectively only 54 kilo binary byte, 0.053 mega binary byte, and 6.67 mJ. Actual gear edge diagnosis experiments confirmed the effectiveness, which can implement model inference in 0.045 s at the edge fault diagnosis node. It is anticipated that the proposed method will find extensive application in the industrial edge interpretable diagnosis.}, archive = {J_EAAI}, author = {Qihang Wu and Jun Luo and Wenbin Huang and Xiaoxi Ding}, doi = {10.1016/j.engappai.2025.112241}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112241}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A quantized subtraction-convolution network for industrial lightweight edge interpretable diagnosis}, volume = {161}, year = {2025}, }
• (2025). Lightweight multi-level feature integration transformer for image super-resolution. EAAI, 161, 112240. (www)

  @article{ , abstract = {Transformer-based methods have attracted significant attention in the field of image super-resolution. However, existing approaches typically concentrate on a single level of information for image reconstruction, and neglect the critical role of multi-level information in feature representation, resulting in the low utilization of the potential capabilities of Transformer. To address this limitation, we propose a novel Transformer model, named as Multi-Level Differential Window Attention Transformer (MLDAT), designed for multi-level information fusion. Specifically, our approach introduces a self-attention module based on differential windows to comprehensively extract and integrate feature information across varying window sizes. Additionally, we introduce a High-order Global Attention Module (HGAB) to combine the second-order attention with global self-attention, which facilitates the establishment of relationships between local features and the overall global feature context within an image while complementing local window information. Extensive experimental results demonstrate that our model can significantly improve the performance of image super-resolution , and achieves the better results compared with existing methods.}, archive = {J_EAAI}, author = {Shuheng Wang and Ziao Gong and Mengda Li and Shen Wu and Yilin He}, doi = {10.1016/j.engappai.2025.112240}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112240}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight multi-level feature integration transformer for image super-resolution}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional object detection for autonomous driving via deep learning: A review. EAAI, 161, 112238. (www)

  @article{ , abstract = {With the rapid advancement of autonomous driving technology, there is an increasing demand for highly accurate and real-time vehicle perception systems. From an artificial intelligence (AI) perspective, three-dimensional (3D) object detection benefits from recent progress in deep neural networks, convolutional neural networks (CNNs), and transformer architectures, which provide powerful tools for feature extraction, spatial reasoning, and multi-modal data fusion. These AI techniques enable robust, uncertainty-aware predictions by effectively modeling complex sensor data. From an engineering standpoint, 3D object detectors serve as critical components in autonomous driving systems by translating AI-derived insights into real-time, high-accuracy vehicle perception. This paper reviews the research progress of deep learning-based 3D object detection algorithms in autonomous driving. First, commonly used data acquisition sensors are systematically categorized, and the most widely adopted 3D detection datasets and evaluation metrics are introduced; the standard 3D bounding-box representation and core network architectures are also explained. Second, algorithms are classified according to input data type: 1) single-modal methods, including vision-based, 3D data dimensionality reduction, point cloud-based, transformer-based, mamba-based, and hybrid point cloud approaches; 2) multi-modal methods, subdivided into serial fusion and parallel fusion strategies within network pipelines. The characteristics, contributions, and limitations of each category are summarized, and representative algorithms are compared across datasets to identify research trends. Finally, current challenges, such as sparse data handling, domain shifts, and real-time constraints, are examined, and prospective directions for future development are proposed.}, archive = {J_EAAI}, author = {Jiaqi Cai and Yiquan Wu}, doi = {10.1016/j.engappai.2025.112238}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112238}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional object detection for autonomous driving via deep learning: A review}, volume = {161}, year = {2025}, }
• (2025). Unscented kalman filter neural network with double-layer decomposition algorithm applied to the prediction of current efficiency in aluminum electrolysis processes. EAAI, 161, 112237. (www)

  @article{ , abstract = {Predictive modeling in predictive optimization control technology can effectively reduce energy consumption and improve production efficiency in the electrolytic aluminum process (EAP). Among the various modeling approaches, artificial neural networks (ANN) have been widely adopted in the EAP due to their strong capability to capture nonlinear relationships inherent in complex industrial systems. However, conventional ANN heavily rely on historical data to achieve optimal models, which limits their accuracy and generalization performance under strong disturbances and time-varying conditions. To address these problems, this article proposes a novel dynamic prediction model of unscented Kalman filter neural network with double-layer decomposition (UKFNN-DD), building upon the foundation of Kalman filter neural network (KFNN). First, singular value decomposition (SVD) is adopted to compute the square root of covariance matrices, enhancing the numerical robustness of the prediction algorithm and overcoming the shortcomings of traditional Cholesky decomposition. Furthermore, a dual-layer KFNN strategy is introduced to overcome the absence of one-step-ahead prediction in conventional state-space formulations. By applying a two-stage correction to the state variables using measurement data, the proposed method improves the adaptability of the model to external environmental variations. Finally, the prediction error of the state variables is optimized using a gradient descent algorithm, which improves the stability and reliability of the model’s prediction performance. Experimental results demonstrate that the proposed method significantly outperforms baseline methods, achieving a 4.38-fold reduction in mean absolute error (MAE) and a 77.29-fold reduction in the sum of squared errors (SSE), thereby verifying its superiority in dynamic prediction for aluminum electrolysis.}, archive = {J_EAAI}, author = {Xiaoyan Fang and Xihong Fei and Zhenyi Xu and Lei Su and Jing Wang}, doi = {10.1016/j.engappai.2025.112237}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112237}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unscented kalman filter neural network with double-layer decomposition algorithm applied to the prediction of current efficiency in aluminum electrolysis processes}, volume = {161}, year = {2025}, }
• (2025). Feature super-resolution-based method for small-scale target detection and segmentation in side-scan sonar. EAAI, 161, 112235. (www)

  @article{ , abstract = {The application of side-scan sonar in underwater target detection plays a significant role in marine engineering construction and ocean resource exploration. However, existing methods for small object detection often suffer from limited accuracy and robustness. To address this issue, we propose a feature super-resolution-based approach for detecting and segmenting small targets in side-scan sonar images. During network training, a feature super-resolution branch is first constructed. Both low-level and high-level features from the backbone of the You Only Look Once (YOLO) network are fed into this branch. After processing through an encoder-decoder architecture, a super-resolved feature map is reconstructed, and the network is optimized via backpropagation to enhance the ability to extract features of small targets. Furthermore, an exponential decay strategy is adopted to define the loss weights of different branches, establishing a branch-aware training mechanism to improve training effectiveness. During inference, the super-resolution branch is discarded to balance detection accuracy and inference efficiency. Experimental results demonstrate that the proposed method achieves superior performance in detecting and segmenting small-scale targets in side-scan sonar imagery, achieving state-of-the-art results on two public side-scan sonar small object datasets. Additionally, this approach can be extended as a training strategy for side-scan sonar target detection and segmentation networks. The source code is available at https://github.com/Yang-Code984/FSR_Sonar .}, archive = {J_EAAI}, author = {Zhiwei Yang and Jianhu Zhao and Xi Zhao and Chao Huang}, doi = {10.1016/j.engappai.2025.112235}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112235}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature super-resolution-based method for small-scale target detection and segmentation in side-scan sonar}, volume = {161}, year = {2025}, }
• (2025). Data-driven prediction of thermal and flow fields in magnetized boger-micropolar tri-hybrid nanofluids via deep artificial neural networks. EAAI, 161, 112232. (www)

  @article{ , abstract = {This study presents a deep learning framework for predicting the thermal and flow behavior of Boger micropolar tri-hybrid nanofluids under magnetized axial squeezing flow between two parallel disks. The fluid comprises gold, silver, and diamond nanoparticles dispersed in a water-based solution, forming a high-performance ternary hybrid heat transfer medium. A fully connected artificial neural network with three hidden layers (32–16–8 neurons) and hyperbolic tangent activation functions was trained on synthetic data generated using a boundary value problem solver. The model predicts four key physical quantities: axial velocity, streamwise velocity, microrotation, and temperature profile. Quantitative evaluation reveals excellent agreement between machine learning predictions and numerical benchmarks, with mean absolute errors consistently below 0.02 and coefficients of determination exceeding 0.99 across all outputs. Sensitivity analysis reveals the impact of the solvent fraction and vortex viscosity parameters on flow penetration and thermal stratification near the boundaries. This work introduces a novel combination of Boger–micropolar fluid dynamics, magnetohydrodynamic squeezing flow, and tri-hybrid nanoparticles into a unified ANN (Artificial Neural Networks) based surrogate, enabling accurate multi-output prediction of strongly coupled nonlinear fields. The proposed approach offers a scalable machine learning model for real-time optimization of nanofluid-based thermal management systems in engineering applications.}, archive = {J_EAAI}, author = {Mohammad Jalili and Hesam Ehsani and Ali Mirzagoli Ganji and Amirali Shateri and Mehdi mahboobtosi and Yuping Wu and Payam Jalili and Bahram Jalili and Davood Domiri Ganji}, doi = {10.1016/j.engappai.2025.112232}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112232}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven prediction of thermal and flow fields in magnetized boger-micropolar tri-hybrid nanofluids via deep artificial neural networks}, volume = {161}, year = {2025}, }
• (2025). Explainable knowledge recommendation for product innovation concept design based on knowledge graph and multi-task learning. EAAI, 161, 112231. (www)

  @article{ , abstract = {In the process of product innovation concept design, by enhancing the transparency of the knowledge recommendation process and providing designers with explainable recommendation results, decision-making time can be reduced, and design efficiency can be improved. Existing explainable knowledge recommendation methods can generate textual explanation information, but they often overlook the dynamic nature of the design process, which negatively affects the accuracy of the recommendations. To address this issue, the current study proposes a knowledge recommendation method for product innovation concept design based on knowledge graph (KG) and multi-task learning. Specifically, a TransD-based model is first constructed to perform the knowledge graph embedding (KGE) task. Then, KGE and Gated Recurrent Unit (GRU) are used to obtain dynamic knowledge demands across different temporal scales from the historical interactions of designers to enhance recommendation accuracy. A multi-task learning framework is subsequently introduced to enable feature sharing between the two tasks, improving the generalization and effectiveness of the model. Finally, an explanation strategy is designed by combining embedded knowledge and paths to provide optimal explainability information. Experimental results demonstrate that, compared with other state-of-the-art knowledge recommendation algorithms, the proposed method not only offers explainable recommendation results but also achieves higher accuracy, making it more suitable for real-world recommendation scenarios.}, archive = {J_EAAI}, author = {Yida Hong and Wenqiang Li and Hai Xiang and Chuanxiao Li and Changfu Wan}, doi = {10.1016/j.engappai.2025.112231}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112231}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable knowledge recommendation for product innovation concept design based on knowledge graph and multi-task learning}, volume = {161}, year = {2025}, }
• (2025). Towards reproducible machine learning-based process monitoring and quality prediction research for additive manufacturing. EAAI, 161, 112223. (www)

  @article{ , abstract = {Additive manufacturing (AM) is increasingly adopted across industries for its ability to support design flexibility, rapid prototyping, and mass customization. Machine learning (ML)-based cyber-physical systems (CPSs) have been extensively developed to improve the print quality of AM. However, the reproducibility of these systems has not been thoroughly investigated due to a lack of formal evaluation methods. Reproducibility, a critical component of trustworthy artificial intelligence, is achieved when an independent team can replicate the findings or artifacts of a study using a different experimental setup and achieve comparable performance. In many publications, critical information necessary for reproduction is often missing due to a lack of comprehensive AM and ML domain knowledge, resulting in systems that fail to replicate the reported performance. Integrating AM and ML domain knowledge, this paper proposes a reproducibility investigation pipeline and a reproducibility checklist for ML-based AM process monitoring and quality prediction systems. Based on the CRoss Industry Standard Process (CRISP) methodology, the pipeline guides researchers through the key steps required to reproduce a study, while the checklist systematically extracts information relevant to reproducibility from the publication. We validated the proposed approach through two case studies: reproducing a fused filament fabrication warping detection system and a laser powder bed fusion melt pool area prediction model. Both case studies confirmed that the pipeline and checklist successfully identified missing information, improved reproducibility, and enhanced the performance of reproduced systems. Based on the proposed checklist and leveraging large language models, a reproducibility survey was conducted to assess the current reproducibility status within this research domain.}, archive = {J_EAAI}, author = {Jiarui Xie and Mutahar Safdar and Andrei Mircea and Bi Cheng Zhao and Yan Lu and Hyunwoong Ko and Zhuo Yang and Yaoyao Fiona Zhao}, doi = {10.1016/j.engappai.2025.112223}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112223}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards reproducible machine learning-based process monitoring and quality prediction research for additive manufacturing}, volume = {161}, year = {2025}, }
• (2025). Degradation-based predictive energy management for intelligent fuel cell hybrid electric vehicles with a novel deep reinforcement learning architecture. EAAI, 161, 112222. (www)

  @article{ , abstract = {The degradation of fuel cell systems (FCS) affects the energy management performance of fuel cell hybrid electric vehicles (FCHEVs), especially in the case of severe degradation. This study develops a novel predictive energy management architecture, which consists of the Extended Long Short-Term Memory (xLSTM) and Soft Actor-Critic (SAC). Specifically, the speed predictor is built using xLSTM network and leverages the speed and position information of the preceding and following vehicles. In order to provide a reliable basis for energy management in degradation scenarios, an FCS degradation model is developed, which enables the dynamic mapping of output efficiency curves under varying state-of-health (SOH) conditions. Sequentially, a SAC-based agent is employed as the energy management strategy (EMS), which innovatively takes full account of the impact of the FCS SOH on energy allocation, achieving proactive degradation-aware energy allocation. Simulation results demonstrate that the developed xLSTM architecture achieves a 71 % improvement in speed prediction accuracy compared to Transformer-based models within the Next Generation Simulation validated scenario. Moreover, at SOH = 90 %, the newly designed EMS reduces operational costs by 8.7 % and 10.7 % compared to conventional methods under the New European Driving Cycle and Worldwide Harmonized Light Vehicles Test Procedure, while decreasing FCS degradation costs by 17 % and 51 %, respectively. The innovative approach not only elevates energy efficiency but also prolongs FCS operational longevity via intelligent SOH-aware, which establishes a new paradigm for lifecycle-optimized energy management in FCHEVs.}, archive = {J_EAAI}, author = {Zhigen Nie and Zhuangfeng Shi and Yufeng Lian and Hao Song}, doi = {10.1016/j.engappai.2025.112222}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112222}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Degradation-based predictive energy management for intelligent fuel cell hybrid electric vehicles with a novel deep reinforcement learning architecture}, volume = {161}, year = {2025}, }
• (2025). Observer based fault tolerant control design for saturated nonlinear systems with full state constraints via a novel event-triggered mechanism. EAAI, 161, 112221. (www)

  @article{ , abstract = {Safety plays a crucial role in many promising applications as they must comply with stringent safety regulations while staying within physical limits. Therefore, safety fault-tolerant systems with saturated nonlinearities and input constraints are investigated in this paper. An event-triggered safety fault tolerant control (FTC) method based on adaptive dynamic programming (ADP) is proposed. The constrained state is modeled using a smooth mapping function, thus transforming the original system into an unconstrained framework. A fault observer is designed for unknown actuator faults occurring in the system. When an unknown fault occurs in the system, the actuator is compensated in real time according to the fault. The optimal safety value function is approximated by constructing a single critic neural network (NN). Then, a novel event-triggered mechanism is proposed, which allows the algorithm to obtain the optimal control law without constructing the event-triggered Hamilton–Jacobi–Bellman (HJB) equation. In addition, by adjusting the size of the parameter τ under the event triggered condition, different demands on the number of event triggers are realized, which in turn ensures the efficient use of resources while leading to the trade-off of optimal control. This paper also incorporates an empirical playback technique in the design of the critic’s update law to address the challenges associated with continuous incentive requirements. It is theoretically proven that all signals of the system are consistent with the limit bounds, thus ensuring the stability of the system.}, archive = {J_EAAI}, author = {Chunbin Qin and Mingyu Pang and Zhongwei Wang and Suyang Hou and Dehua Zhang}, doi = {10.1016/j.engappai.2025.112221}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112221}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Observer based fault tolerant control design for saturated nonlinear systems with full state constraints via a novel event-triggered mechanism}, volume = {161}, year = {2025}, }
• (2025). Cybersecurity enhancement using conditional generative adversarial network with transformer-based conditional variational autoencoder. EAAI, 161, 112220. (www)

  @article{ , abstract = {Since, Artificial Intelligence is highly developing and concatenating into several domains, cybersecurity is an important field of delivering both the advantages and disadvantages. In addition to this, Artificial Intelligence is applied in a wide variety of applications like healthcare sector, content creation and entertainment and financial industries. Therefore, this work finds the efficiency of Artificial Intelligence -oriented cybersecurity metrics in succeeding the digital environment over elevating cyber threats. Here, the developed models consist of two different stages while implementing the model. In the first stage, the essential dataset is assembled from the benchmark data source. These datasets are assembled by using Generative Artificial Intelligence (Gen Artificial Intelligence networks). Consequently, the raw data is given as an input to Conditional Hybrid Network for cybersecurity enhancement. Further, the Transformer-based Conditional Variational Autoencoders with Spatial-temporal Attention are designed for feature extraction that is subjected to the Conditional Generative Adversarial Network for classifying the cyber attacks. Henceforth, the developed network is evaluated and designed with multiple measures. Comparing baseline models, the suggested network obtains higher performance for developing security over cyber networks.}, archive = {J_EAAI}, author = {Prithvipal Singh and Sandeep Singh and Gurupdesh Singh and Amritpal Singh}, doi = {10.1016/j.engappai.2025.112220}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112220}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cybersecurity enhancement using conditional generative adversarial network with transformer-based conditional variational autoencoder}, volume = {161}, year = {2025}, }
• (2025). A modal transfer enhanced deep learning for structural dynamic response with sparse spatial data. EAAI, 161, 112218. (www)

  @article{ , abstract = {Sparse spatial data in structural health monitoring (SHM) presents significant challenges due to large spatial gaps and high uncertainty, which hinder the generalization ability and performance of traditional data-driven methods that rely on dense sensor networks. To address this issue, this paper proposes a novel modal transfer-enhanced deep learning (MT-DL) model for the reconstruction and prediction of structural dynamic responses. The core idea is to integrate physical modal information (mode shapes) into the deep learning framework via transfer learning. This integration enables the model to capture the spatial correlation among structural nodes, even under extremely limited data conditions. The MT-DL model is initially validated on a simply supported beam with varying stiffness, where it demonstrates significantly higher reconstruction accuracy compared to both conventional deep learning (DL) approaches and the graph convolutional network (GCN) model under sparse sensor conditions. To demonstrate the model's robustness and generalizability, it is further applied to the dynamic response reconstruction of a continuous beam subjected to moving loads, a plate with complex boundary conditions under impact loading, and a slender flexible riser undergoing vortex-induced vibration (VIV). The results show that the proposed MT-DL model, by incorporating physical principles into data-driven learning, not only enhances prediction accuracy with minimal data but also offers improved physical interpretability. This approach provides a promising solution for structural monitoring in scenarios where dense sensor instrumentation is impractical or cost-prohibitive.}, archive = {J_EAAI}, author = {Yangyang Liao and Yajuan Xie and Hesheng Tang and Zihan Xia and Songtao Xue}, doi = {10.1016/j.engappai.2025.112218}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112218}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A modal transfer enhanced deep learning for structural dynamic response with sparse spatial data}, volume = {161}, year = {2025}, }
• (2025). A novel deep learning pipeline for coronary artery analysis in X-ray angiography. EAAI, 161, 112217. (www)

  @article{ , abstract = {X- ray angiography is a primary and essential assistive method for the accurate diagnosis of coronary blockage. Segmentation of coronary arteries and analysis of coronary artery blockage are the two primary tasks performed by cardiologists. We have proposed Coronary Artery Segmentation, Blockage Detection, and Measurement (CASBloDaM) a state-of-the-art pipeline integrated deep learning framework with conventional image processing techniques to achieve accurate coronary artery analysis in X-ray angiographic images. A new private dataset of 214 X-ray angiographic images is prepared for training, validation, and testing of the model by manual pixel annotation. The UNet3+ deep learning model is trained on X-ray angiographic images for accurate artery segmentation while multiple image processing techniques are developed to perform blockage detection and measurement. The model shows excellent performance in artery segmentation and the dice score of 0.989, accuracy of 0.985, specificity of 0.913 and sensitivity of 0.847 is achieved during the testing which is superior when compared against the previous reported works. An accuracy of 0.853, 0.882 and 0.725 are achieved by the proposed catheter detection, sub-artery removal and blockage detection algorithms respectively when evaluated for 150 test images. The proposed methodology achieved a Mean Square Error (MSE) of 28.66 and an R-square of 0.81 for blockage percentage estimation, and an MSE of 69.91 with an R-square of 0.99 for blockage location, compared to manual calculations. The novelty of the present work is the end-to-end integrated framework which can accurately perform coronary artery analysis in X-ray angiographic images and propose its clinical usage.}, archive = {J_EAAI}, author = {Karan V. Padariya and Abhishek Raval and Pranay Soni and Harsh Kapadia}, doi = {10.1016/j.engappai.2025.112217}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112217}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel deep learning pipeline for coronary artery analysis in X-ray angiography}, volume = {161}, year = {2025}, }
• (2025). A multi-teacher knowledge distillation-based framework for long-term respiratory monitoring and prediction with a novel flexible wearable sensor in healthcare engineering. EAAI, 161, 112216. (www)

  @article{ , abstract = {Respiratory monitoring plays a critical role in early health warnings and preventive care, offering significant potential for advancements in healthcare engineering. Wearable respiratory monitoring devices, known for their compact design, portability, and real-time capabilities, face challenges such as limited long-term comfort, environmental interference, and signal inaccuracies. In this study, we propose a novel wearable respiratory monitoring framework, RAMP, which integrates an innovative artificial muscle-based flexible sensor system with advanced deep learning modules. The system is designed to reconstruct and analyze respiratory data across various human activities and predict long-term respiratory function. Utilizing a multi-teacher knowledge distillation mechanism, the framework optimizes a student model for enhanced prediction accuracy. Experimental results demonstrate the mean absolute percentage error (MAPE) of 7.28 and mean absolute error (MAE) of 11.92, highlighting the feasibility and effectiveness of system. This work advances the development of portable health monitoring devices and provides a robust foundation for long-term respiratory activity assessment and forecasting, contributing to the broader field of healthcare engineering and personalized medicine.}, archive = {J_EAAI}, author = {Ke Li and Qing Wang and Haoke Liu and Mingke Wang and Suiyuan Zhu and Xiang Wang and Jing Qin}, doi = {10.1016/j.engappai.2025.112216}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112216}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-teacher knowledge distillation-based framework for long-term respiratory monitoring and prediction with a novel flexible wearable sensor in healthcare engineering}, volume = {161}, year = {2025}, }
• (2025). Dynamic mask network based on spiking neural convolutional model for missing modality brain tumor segmentation. EAAI, 161, 112215. (www)

  @article{ , abstract = {Brain tumor segmentation is a medical image processing task aimed at accurately locating and isolating tumor regions from brain scan images (e.g., Magnetic Resonance Imaging, MRI) in order to help doctors in diagnosis, treatment planning and surgical navigation. Automatic brain tumor segmentation is extremely challenging due to incomplete feature representation in the case of missing modalities and insufficient inter-modal information interaction. To this end, this paper proposes a novel dynamic threshold mask Transformer network for the missing modality brain tumor segmentation task, which is designed based on the nonlinear spiking neural convolutional model. The network consists of four independent encoders and a shared decoder to extract the features of each modality and perform shared representation learning. Among them, the dynamic threshold mask Transformer introduces learnable embedding vectors, generates dynamic masks on top of static masks to achieve fine-grained feature filtering, and enhances the ability of inter-modal information interaction. The adaptive gating weighting module and the channel cross spiking neural P attention module fuse modal features layer by layer in both spatial and channel dimensions to strengthen the modeling capability of local and global features. We conducted extensive comparative experiments on different missing modal cases in the BraTS2020 and BraTS2018 datasets. The experimental results show that the method effectively improves the robustness of missing modalities and the performance of brain tumor segmentation while maintaining the computational efficiency, and has good generalization ability and practicality.}, archive = {J_EAAI}, author = {Junjie Li and Rui Cai and Bing Li and Hong Peng}, doi = {10.1016/j.engappai.2025.112215}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112215}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic mask network based on spiking neural convolutional model for missing modality brain tumor segmentation}, volume = {161}, year = {2025}, }
• (2025). Enhancing accuracy in fall detection and prediction for elderly individuals using ensemble wavelet neural network and maximal overlap discrete wavelet transform. EAAI, 161, 112213. (www)

  @article{ , abstract = {Falls are the second-largest risk factor for the health of the elderly. Various researchers have proposed approaches for monitoring health and falls using only the Internet of Things (IoT) and sensors. However, relying on a single sensor limits accuracy in fall risk prediction. Data from multiple sensors can be used to classify human posture through learning-based algorithms stemming from machine learning. Recently, posture-detecting sensors have become increasingly popular, with InvenSense's Inertial Measurement Unit 9250 (IMU 9250) being a notable example for detecting elderly motion and transmitting data to a computer. This paper proposes the Ensemble Wavelet Network Framework (EWNNET) for fall prediction and detection using the Maximal Overlap Discrete Wavelet Transform (MODWT). The inertial measurement unit (IMU) dataset, containing real-time posture data from seven sensors, is split into 80 % for training and 20 % for testing. The EWNNET approach is compared with other machine-learning methods for fall detection, showing improved accuracy, sensitivity, precision, F-score, and specificity, with a 96.7 % classification success rate.}, archive = {J_EAAI}, author = {Safa Hussein Mohammed and Yangyu Fan and Guoyun Lv and Shiya Liu}, doi = {10.1016/j.engappai.2025.112213}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112213}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing accuracy in fall detection and prediction for elderly individuals using ensemble wavelet neural network and maximal overlap discrete wavelet transform}, volume = {161}, year = {2025}, }
• (2025). Leakage localization methodology based on dynamic pressure signal for subsea pipeline. EAAI, 161, 112212. (www)

  @article{ , abstract = {Leakage is one of the most critical failure forms of subsea pipeline. The accurate leakage localization is of great significance to ensure the safe and reliable transportation of subsea pipeline. Leakage locations are considered to be discretely distributed along the subsea pipeline. However, an overabundance of nodes in leakage localization model is caused by excessive discretization. The performance of a leakage localization model with much localization points is poor. Furthermore, the data collected in site usually contains a lot of noise which reduce the effectiveness of leakage localization. A leakage localization methodology based on dynamic pressure signal for subsea pipeline is proposed in this paper. A noise reduction model based on variational mode decomposition (VMD) algorithm combined with power spectral density (PSD) is established to reduce noise of leakage signal. An improved K-means grouping model is developed to mining data for inherent similarity and cluster leakage characteristic. It improves robustness of the leakage localization model. A radial basis function (RBF) neural network leakage localization model optimized by the pelican optimization algorithm (POA) is used to identify the location of leakage. A leakage experiment is used to study performance of this methodology. The localization accuracy of the proposed leakage localization methodology is more than 90 %, the localization error is less than 16 cm. After neural network combined with improved grouping model, average leakage localization accuracy increased by 15.65 %, average absolute error decreased 8.64 cm. The proposed methodology provides an effective tool for leakage localization of critical equipment in subsea production system.}, archive = {J_EAAI}, author = {Guowei Ji and Baoping Cai and Xuelin Liu and Yi Jiang and Yixin Zhao and Qingping Li and Lei Gao and Kaizheng Wu}, doi = {10.1016/j.engappai.2025.112212}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112212}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leakage localization methodology based on dynamic pressure signal for subsea pipeline}, volume = {161}, year = {2025}, }
• (2025). Enhancing entity and relation extraction with dynamic hard negative augmentation framework. EAAI, 161, 112211. (www)

  @article{ , abstract = {Entity and relation extraction, a fundamental task in information extraction, plays a crucial role in modeling unstructured text by identifying meaningful entities and their semantic relationships. While existing methods have shown effectiveness, they still face challenges in accurately identifying entity boundaries and extracting complex relationships. These challenges primarily arise from current contrastive learning approaches, which uniformly handle all negative samples in boundary detection and relation extraction without emphasizing the learning of hard negative samples. Additionally, the scarcity of hard negative samples limits the exploration of the state space near the anchors. To tackle these challenges, we introduce Dynamic Hard Negative Augmentation, an innovative framework designed to strategically explore and generate hard negative samples, thereby enhancing the learning of challenging cases through adaptive contrastive learning. During the negative sample augmentation process, we employ adversarial training to explore underrepresented areas of hard negative samples, generating a comprehensive coverage of the hard negative sample space to effectively explore the state space. We further introduce a dynamic enhancement mechanism that continuously optimizes the proportion of hard negative samples during training, ensuring targeted learning of these challenging cases.}, archive = {J_EAAI}, author = {Qibin Li and Shengyuan Bai and Nai Zhou and Nianmin Yao}, doi = {10.1016/j.engappai.2025.112211}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112211}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing entity and relation extraction with dynamic hard negative augmentation framework}, volume = {161}, year = {2025}, }
• (2025). NashDQNSleep: Nash-based deep Q-network adaptive sleep scheduling for energy efficiency and age of information optimization in industrial internet of things. EAAI, 161, 112210. (www)

  @article{ , abstract = {Fresh and timely data is essential for the sustainable operation of Industrial Internet of Things (IIoT) systems, which support real-time monitoring and decision-making tasks. Sensor nodes typically need to remain active for long durations to maintain data freshness, resulting in high energy consumption. Traditional sleep scheduling methods often trade off energy savings with data freshness, and many rely on centralized sink-node coordination, which can create communication bottlenecks and increase energy use. This paper introduces NashDQNSleep , a novel decentralized sleep scheduling algorithm that uniquely combines game-theoretic modeling with deep reinforcement learning to address these challenges. Unlike conventional DQN approaches that optimize node behavior independently or centrally, NashDQNSleep formulates the scheduling problem as a general-sum stochastic game and uses Deep Q-Networks (DQNs) to compute Nash equilibria. This enables sensor nodes to make autonomous, yet strategically coordinated decisions based on peer-to-peer information exchange, achieving a stable and efficient balance between energy consumption and data freshness without centralized control. Extensive simulations on diverse IIoT network topologies demonstrate that NashDQNSleep improves energy efficiency by up to 35%, reduces Age of Information (AoI) by 40%, and increases successful data transmissions by 15%–20% compared to state-of-the-art decentralized and centralized methods. These results establish NashDQNSleep as an effective, scalable, and practical solution for reliable and sustainable IIoT operations.}, archive = {J_EAAI}, author = {Partha Sarathi Banerjee and Soumyapriya Goswami and Debashis De}, doi = {10.1016/j.engappai.2025.112210}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112210}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {NashDQNSleep: Nash-based deep Q-network adaptive sleep scheduling for energy efficiency and age of information optimization in industrial internet of things}, volume = {161}, year = {2025}, }
• (2025). A novel contrastive learning framework for multi-parameter optimization in 3D printing. EAAI, 161, 112209. (www)

  @article{ , abstract = {Additive manufacturing (3D printing) revolutionizes prototyping and production through unparalleled material efficiency. However, part quality remains highly sensitive to parameter variations, where subtle deviations induce defects, material waste, and process instability. Traditional quality control methods – reliant on rule-based heuristics or manual inspections – fail to address complex multi-parameter interactions and fine-grained anomalies, limiting industrial scalability. This article focuses on an important issue in 3D printing: How can we develop a robust, automated framework to simultaneously detect and classify subtle multi-parameter anomalies in 3D printing, overcoming the limitations of manual and single-defect-focused approaches? We propose a supervised contrastive learning framework integrating Vision Transformers (ViT) to learn discriminative feature representations for multi-parameter optimization. By maximizing intra-class similarity and inter-class separation, our model captures nuanced variations across printing scenarios. The ViT architecture processes real-time printing images, while contrastive loss ensures compact feature clusters for “Low”, “Optimal”, and “High” parameter classes. Experimental evaluations on open-source datasets demonstrate our framework achieves 8.45% accuracy, outperforming conventional CNNs by 10.11%. Real-world validation shows robust performance across critical parameters: flow rate (86.5% accuracy in nominal ranges), feed rate (87% accuracy), and extrusion temperature (90% accuracy at optimal settings). The ViT’s self-attention mechanism enables precise detection of localized anomalies, such as under-extrusion and layer misalignment.}, archive = {J_EAAI}, author = {Jieyang Peng and Simon Kreuzwieser and Dongkun Wang and Andreas Kimmig and Zhi Fan and Jianing Li and Jivka Ovtcharova}, doi = {10.1016/j.engappai.2025.112209}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112209}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel contrastive learning framework for multi-parameter optimization in 3D printing}, volume = {161}, year = {2025}, }
• (2025). A lightweight model based on multi-scale feature fusion for ultrasonic welding surface defect detection. EAAI, 161, 112208. (www)

  @article{ , abstract = {Ultrasonic welding technology is crucial in industrial and medical fields, relying on precise surface defect detection for quality assurance. Traditional methods suffer from low accuracy, efficiency, high costs, and complex implementation. Additionally, current neural networks for ultrasonic surface defect detection struggle to balance parameter optimization with detection accuracy. To solve this problem, we proposed a lightweight model based on multi-scale feature fusion for the Ultrasonic Weld Surface Defect Detection Network (UWSDNet). First, the feature extraction module with reparameterization technology (FRT) and application of efficient multi-scale attention (EMA) are proposed to alleviate network redundant parameters and computational overhead brought by welding background. Secondly, the multi-core feature enhancement module (MCM) is introduced. It enhances multi-scale object detection with fewer parameters to cope with the actual edge deployment of ultrasonic welding. Finally, the lightweight asymmetric detection head (LADH) and contextual and spatial feature calibration network (CSFCN) are introduced into the network. To improve the multi-core dimensional feature capture capability, to solve the problem of large size span of ultrasonic welding surface defects. Experimental evaluations on a self-built ultrasonic welding wire harness defect dataset show that UWSDNet achieves the mean average precision (mAP) of 88.9%, the precision of 95.6% with parameters of 12.7M. In addition, UWSDNet achieves excellent performance on the publicly available NEU-DET dataset, demonstrating strong generalization and application potential in industrial defect detection.}, archive = {J_EAAI}, author = {Rui Liu and Lun Zhao and Yu Ren and Zhonghua Shen and Liya Li and Jianfeng Luo and Zeshan Abbas}, doi = {10.1016/j.engappai.2025.112208}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112208}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight model based on multi-scale feature fusion for ultrasonic welding surface defect detection}, volume = {161}, year = {2025}, }
• (2025). Artificial neural network solution for magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating jeffery-hamel flow under heat generation: Entropy generation analyses. EAAI, 161, 112207. (www)

  @article{ , abstract = {This research-work examines entropy generation in Magnetohydrodynamic (MHD) ternary hybrid nanofluid under rotating Jeffery-Hamel flow with heat generation alongside thermal radiation. Using appropriate similar transformation equations, the original partial differential equations (PDEs) are converted into nonlinear ordinary differential equations (ODEs), while Artificial Neural Networks (ANNs), which highly extensively used as universal function approximators, is considered in this investigation. In fact, to provide efficient solutions for the nonlinear transformed problem of the magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating Jeffery-Hamel flow, solvers consider log-sigmoid, radial basis and tan-sigmoid activation functions, optimized with an interior point method to generate optimal weights of each considered ANN model. The model relies on assumptions of incompressible steady flow together with negligible viscous energy loss and usage of Rosseland radiation approximation. The predictions made through the Artificial Neural Network (ANN) match numerical benchmarks successfully thus demonstrating better accuracy when modeling velocity, temperature and entropy profiles. The research data indicates that systems attaining better thermal qualities operate effectively under elevated nanoparticle levels and radiation conditions which provides fundamental knowledge to enhance thermal system design. The results from the proposed schemes match numerical solutions precisely thus showing their high accuracy for evaluating ternary hybrid nanofluid flow and heat behavior. Results show that the velocity and temperature profiles of the Magnetohydrodynamic (MHD) ternary hybrid nanofluid reveal significant improvements at higher thermal volume fractions of nanoparticles. The rate of flow increases, perhaps because inertial forces prevail over viscous forces, thereby improving flow dynamics. With higher Radiation numbers associated with higher permeability, the local temperature gradients decrease promoting heat transfer rates within the fluid.}, archive = {J_EAAI}, author = {Nouar Ahcene and Amar Dib and Farhan Lafta Rashid and Kezzar Mohamed and Mohamed Rafik Sari and Manal Elzain and Hamiden Abd El-Wahed Khalifa}, doi = {10.1016/j.engappai.2025.112207}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112207}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural network solution for magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating jeffery-hamel flow under heat generation: Entropy generation analyses}, volume = {161}, year = {2025}, }
• (2025). Intelligent compound fault decoupling of rolling bearing based on parallel capsule network. EAAI, 161, 112206. (www)

  @article{ , abstract = {Due to the complex working environment of rolling bearings, various components such as inner ring, outer ring, rolling element and cage in bearing may interact with each other, leading to a compound fault formed by a variety of single fault coupling. Most methods generally regard compound faults as single faults, ignoring the interaction between single faults, which is not conducive to decoupling compound faults into multiple single faults and formulating maintenance plans. Moreover, the capsule network requires stacking multiple capsule layers to enhance performance, which significantly increases model parameters and consumes substantial memory resources. Therefore, a compound fault intelligent decoupling method based on parallel capsule network combining dynamic routing and attention routing is proposed in this study. Firstly, the Omni-Scale block is added to the feature extraction part, which can cover different sizes of receptive fields to enhance the feature extraction ability of the network. Secondly, an attention routing module is proposed, which realize the transmission of information from low-level capsules to high-level capsules by calculating the correlation between the same layers. Finally, the parallel capsule decoupling layer is constructed by using dynamic routing and attention routing. This method is especially suitable for practical engineering scenarios where compound bearing fault samples are limited and computational resources are constrained, providing a lightweight and effective solution for intelligent fault diagnosis. Experimental results show that the proposed method significantly reduces model complexity while maintaining high diagnostic performance under small-sample conditions, with the ablation study further confirming the meaningful contribution of each core module.}, archive = {J_EAAI}, author = {Renwang Song and Chenyu Jiao and Hui Shi and Linying Chen}, doi = {10.1016/j.engappai.2025.112206}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112206}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent compound fault decoupling of rolling bearing based on parallel capsule network}, volume = {161}, year = {2025}, }
• (2025). Deep reinforcement learning-based strategic bidding in electricity markets via variational autoencoder-assisted competitor behavior learning. EAAI, 161, 112205. (www)

  @article{ , abstract = {In a deregulated electricity market, self-interested producers have incentives to offer strategically for maximizing their own profits. While deep reinforcement learning (DRL) has shown great potential for solving such strategic bidding problems, existing methods typically oversimplify strategic action spaces and neglect the influence of competitors' offering behaviors. To bridge these gaps, this paper proposes a novel DRL-based framework to model and solve the strategic bidding problem of an individual producer by jointly considering price-quantity offering actions and the dynamic behaviors of market competitors. First, a bilevel optimization model is formulated to incorporate offering actions on price-quantity pairs. Then, a data-driven framework that combines a variational autoencoder with a density-based clustering method is proposed to learn and capture competitors' offering behaviors. Finally, an imitation learning-integrated DRL algorithm is developed to improve learning stability and solution quality for strategic bidding with price-quantity actions and competitors' offering behaviors. Case studies on the IEEE-30 bus system show that the proposed framework obtains a 28.12 k$ (24.25 %) increase in average profit compared to the existing approach, demonstrating its effectiveness and adaptability under dynamic market conditions.}, archive = {J_EAAI}, author = {Fei Hu and Yong Zhao and Yaowen Yu and Yuanzheng Li}, doi = {10.1016/j.engappai.2025.112205}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112205}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning-based strategic bidding in electricity markets via variational autoencoder-assisted competitor behavior learning}, volume = {161}, year = {2025}, }
• (2025). A deep-supervised framework utilizing edge features and global texture features for distortion rectification in fisheye images. EAAI, 161, 112204. (www)

  @article{ , abstract = {Fisheye cameras, with their ultra-wide field of view, are increasingly deployed in urban environments. However, the severe radial distortion inherent in fisheye imagery presents formidable challenges for downstream computer vision tasks, including object recognition, motion estimation, and semantic segmentation. To address this issue, a deeply supervised framework for distortion rectification that leverages edge features and global texture information (DR-EGT) is proposed, which uniquely integrates edge-aware structural features with global texture information under a unified deep supervision strategy. Unlike conventional approaches that rely solely on texture priors or geometric assumptions, DR-EGT introduces a hierarchical supervision mechanism that simultaneously leverages low-level edge contours and high-level texture semantics to guide the distortion correction process. This joint supervision enables the network to learn the distortion flow field of the entire image, which facilitates the reconstruction of geometrically accurate and perceptually sharp undistorted images. On the Places2 dataset, DR-EGT outperforms existing advanced methods, achieving a 12.12 % increase in PSNR (Peak Signal-to-Noise Ratio), 7.44 % improvement in SSIM (Structural Similarity Index Measure), 16.42 % reduction in MAE (Mean Absolute Error), and a 78.17 % gain in FID (Fréchet Inception Distance), demonstrating its superior reconstruction fidelity and perceptual quality. The results demonstrate the ability of DR-EGT not only correct complex fisheye distortion but also suppress post-correction artifacts and visual degradation.}, archive = {J_EAAI}, author = {Yicheng Chen and Shuang Li and Yuhuan Gu and Chang Feng and Changhai Zhai}, doi = {10.1016/j.engappai.2025.112204}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112204}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep-supervised framework utilizing edge features and global texture features for distortion rectification in fisheye images}, volume = {161}, year = {2025}, }
• (2025). Spectrum prior-based and visibility fusion method for underwater image enhancement. EAAI, 161, 112203. (www)

  @article{ , abstract = {When light propagates in water, it undergoes scattering and absorption phenomena, which typically result in haze, high blur, low contrast and color distortion, making it extremely challenging to obtain high-quality images. To address these issues, many existing methods target image enhancement by correcting specific aspects such as color shift or contrast. However, challenges like poor visibility and low-light conditions are often overlooked. In this paper, we proposed a spectrum prior-based and visibility fusion method (SPV) to enhance underwater images in terms of color, contrast, and visibility. Unlike existing methods, SPV complements the advantages of both physical and non-physical models, comprehensively addressing the problems of reduced visual visibility, color distortion, and low contrast caused by low-light environments, thereby significantly improving the overall image quality. We proposed a dehazing module based on spectral information priors, which reliably restores image quality under complex water conditions. Additionally, we introduced a color correction module based on human color perception and employed morphological operations, effectively solving the issues of color shift and unclear contours in underwater images. Furthermore, we proposed a visibility enhancement module based on the fuzzy c-means clustering method to improve image contrast and visibility, particularly under low-light conditions. Finally, through a detail enhancement fusion module, we simultaneously addressed problems related to color shift, low contrast, and low visibility. SPV showed excellent performance in application tests including feature point matching, geometric rotation estimation, and edge detection. Comparative experiments on four real underwater datasets against 14 advanced enhancement methods demonstrated promising results.}, archive = {J_EAAI}, author = {Qifeng Liu and Xin Yan and Lu Shen and Qiang Li}, doi = {10.1016/j.engappai.2025.112203}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112203}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spectrum prior-based and visibility fusion method for underwater image enhancement}, volume = {161}, year = {2025}, }
• (2025). An integrated multi-criteria decision-making approach with unknown weight information under probabilistic dual-hesitant pythagorean fuzzy environment and its application to supplier selection. EAAI, 161, 112202. (www)

  @article{ , abstract = {The preference ranking organization method for enrichment evaluation (PROMETHEE) has been proved to be one of the most effective techniques to rank alternatives of multi-criteria decision-making (MCDM) problems. However, the existing PROMETHEE cannot accurately adjust the representation range of uncertain information. Besides, the weight determination in PROMETHEE heavily relies on the decision matrix of alternatives on the criteria. Moreover, the information aggregation in PROMETHEE models lacks consideration of the interrelationship among criteria. To address the aforementioned shortcomings, this paper introduces a novel MCDM method that integrates the best-worst method (BWM) and PROMETHEE to help the decision-maker (DM) select the optimum alternative under the probabilistic dual-hesitant Pythagorean fuzzy (PDHPF) environment. Firstly, we extend PROMETHEE method to PDHPF scenario, which not only helps the DM depict subjective evaluations, but also provides the DM with a laxer constraint to present decision information. Secondly, the PDHPF power weighted Hamy mean operator (PDHPFPWHM) is utilized to aggregate the preference information of PROMETHEE. Additionally, the BWM method is utilized and extended to the PDHPF environment for acquiring optimal weights of criteria. The high efficiency and consistency of BWM are well-suited for the complex PDHPF environment. Finally, the method is applied to a semiconductor supplier selection case to demonstrate the validity, superiority, and feasibility of the proposed approach.}, archive = {J_EAAI}, author = {Huzhi Xue and Haihua Xie and Butian Zhao and Jun Wang}, doi = {10.1016/j.engappai.2025.112202}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112202}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated multi-criteria decision-making approach with unknown weight information under probabilistic dual-hesitant pythagorean fuzzy environment and its application to supplier selection}, volume = {161}, year = {2025}, }
• (2025). A novel approach for understanding the parking demand for internal access roads in hub car parks. EAAI, 161, 112201. (www)

  @article{ , abstract = {The hubs serve as pivotal nodes within urban transport networks, and the car parks associated with these hubs constitute an integral component. The prevalent practice in China of utilizing internal access roads within car parks for passenger pick-up by online car-hailing vehicles has engendered a novel category of parking demand. However, this development presents a significant challenge to the efficient operation of hub car parks. A thorough analysis and comprehension of the parking demand for internal access roads is essential in devising suitable management strategies to enhance parking efficiency. Initially, this study categorizes the parking demand for internal access roads into three distinct groups, applying parking duration and the number of entries as classification criteria. Subsequently, considering the number of train frequencies, the categorized parking demands are forecasted independently utilizing the Long Short-Term Memory (LSTM) model, complemented by the Shapley Additive exPlanations (SHAP) method for model interpretation. Ultimately, Vector Autoregression (VAR) is employed to investigate the interaction mechanism between the parking demand for parking spaces and the parking demand for internal access roads. The findings indicate that optimal predictive performance is attained when employing a time interval of 15 min and an input step size of 4. Competition between the parking demand for internal access roads and the parking demand for parking spaces primarily occurs within respective categories, underscoring the need for external interventions when these demands are imbalanced.}, archive = {J_EAAI}, author = {Qianyi Hu and Weidong Liu and Chenyu Yan and Chu Zhang and Jun Chen and Changyin Dong}, doi = {10.1016/j.engappai.2025.112201}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112201}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach for understanding the parking demand for internal access roads in hub car parks}, volume = {161}, year = {2025}, }
• (2025). Machine learning for disease-specific prediction of high-cost patients. EAAI, 161, 112200. (www)

  @article{ , abstract = {High-cost patients incur disproportionately high medical expenses, and identifying them proactively is crucial for effective healthcare management. While previous research has focused on identifying high-cost patients based on overall expenditure, there has been a lack of studies analyzing them in the context of specific diseases. This study addressed this gap by leveraging data from the National Health Insurance Service (NHIS) of South Korea, spanning 2015 to 2019, to develop predictive models for identifying these patients. We trained models using data from 880,000 individuals to predict high-cost patients in 2019 using resource-efficient machine learning algorithms such as Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Neural Networks (NN) that minimize computational overhead, with undersampling techniques applied to handle data imbalance. We focused on the six major disease categories that account for the highest medical expenditures in South Korea: diseases of the musculoskeletal system (DMS), circulatory system (DCS), eye and ear (DEA-DEM), digestive system (DDS), genitourinary system (DGS), and respiratory system (DRS). We discovered that disease-specific analyses revealed important predictive factors that were not apparent in aggregate analyses. For example, hemoglobin levels emerged as crucial predictors for DCS, while body mass index (BMI) proved essential for DMS prediction. These findings enhance our understanding of the factors contributing to high medical costs and provide a foundational framework for healthcare providers and policymakers to develop more targeted and effective health management strategies.}, archive = {J_EAAI}, author = {Inwoo Tae and Hyeongwoo Kong and Junghye Lee and Yongjae Lee}, doi = {10.1016/j.engappai.2025.112200}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112200}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning for disease-specific prediction of high-cost patients}, volume = {161}, year = {2025}, }
• (2025). Optimizing mean conditional value-at-risk portfolios through deep neural network stock prediction. EAAI, 161, 112198. (www)

  @article{ , abstract = {Portfolio optimization is essential in financial decision-making, requiring a balance between risk minimization and return maximization. Effective stock selection significantly influences portfolio performance. Traditional methods often struggle to effectively integrate advanced risk assessment techniques with stock selection. To enhance portfolio diversification and improve risk-adjusted returns, this study integrates deep learning-based stock selection with the mean conditional Value-at-Risk (MCVaR) model and entropy constraints to enhance portfolio diversification and risk-adjusted returns. Various deep neural networks, including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Multi-Layer Perceptrons (MLP), and Radial Basis Function Neural Networks (RBFN), are employed to rank stocks based on risk and return characteristics. The top-ranked stocks with the lowest risk are selected for portfolio construction. The entropy constraint is introduced to prevent excessive weight concentration, ensuring a well-diversified portfolio. Historical datasets from the Bombay Stock Exchange (BSE), India, B3 Stock Exchange, Brazil, and Shanghai Stock Exchange, China, are used for validation, with performance assessed on an out-of-sample dataset. Additionally, the efficacy of the suggested approach is evaluated by contrasting it with other machine learning and conventional portfolio optimization techniques. Experimental results demonstrate that the LSTM+MCVaR model with entropy constraint consistently outperforms other deep learning and conventional optimization methods, achieving superior cumulative returns and Sharpe ratios. The findings highlight the potential of combining LSTM forecasting with MCVaR optimization and entropy regularization for robust, diversified portfolio construction.}, archive = {J_EAAI}, author = {Jyotirmayee Behera and Pankaj Kumar}, doi = {10.1016/j.engappai.2025.112198}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112198}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing mean conditional value-at-risk portfolios through deep neural network stock prediction}, volume = {161}, year = {2025}, }
• (2025). Design and analysis of an unbiased intelligent recommendation system for all-rounders in cricket based on multiple criteria decision making. EAAI, 161, 112197. (www)

  @article{ , abstract = {Designing and analyzing an unbiased intelligent all-rounder recommendation system in cricket is a critical and complex decision-making task, where performance prediction itself is a crucial issue. The majority of existing works on cricket focus on batsmen, bowlers, and teams. However, performance analyses of all-rounders are hardly found. Hence, the motivation of this paper is to propose an artificial intelligence (AI)-based method for assessing the performance of all-rounders utilizing various multiple-criteria decision-making (MCDM) techniques. With this goal in mind, effective attributes are considered for evaluating all-rounder bowling and batting performances. Case studies using a set of 20 all-rounders have been taken from the recent International Cricket Council (ICC) all-rounders list to determine their ranking. The performance of various MCDM techniques is investigated using ICC rankings, with the Spearman Rank Correlation Coefficient applied to the One Day International (ODI) format. The results indicate that the Criteria Importance Through Intercriteria Correlation (CRITIC)-VlseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) method yields quite promising insights, achieving a higher correlation with ICC rankings (Spearman Rank Correlation: 0.794) and hence can be used as an intelligent AI-based all-rounder recommendation system. Finally, we have also conducted a sensitivity analysis on the ranking outcomes of all-rounders to examine the utility and robustness of our proposed MCDM approach. The proposed approach is beneficial for team selectors, analysts, and fantasy sports platforms, offering a fairer and more reliable ranking of all-rounders.}, archive = {J_EAAI}, author = {Nayan Ranjan Das and Imon Mukherjee and Goutam Paul}, doi = {10.1016/j.engappai.2025.112197}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112197}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design and analysis of an unbiased intelligent recommendation system for all-rounders in cricket based on multiple criteria decision making}, volume = {161}, year = {2025}, }
• (2025). Semantic analysis-based recommender system using sequential clustering and convolutional neural network. EAAI, 161, 112196. (www)

  @article{ , abstract = {Accurate prediction of user preferences and generation of personalized recommendations remain as critical challenges in intelligent recommendation systems. In this study, we propose a novel recommendation model that transforms the rating prediction problem into a single-label multiclass classification task. The model integrates three key components: (1) ordered clustering information derived from user review text similarity, (2) rating rank similarity reflecting users’ behavioral tendencies, and (3) a convolutional neural network (CNN) to extract semantic representations from user textual data. First, user review embeddings are clustered to capture high-level semantic preferences, where cluster indices are utilized as ordered categorical features. Second, rating rank similarity features are constructed by comparing the relative ranking of items rated by similar users. These features are fused and fed into a CNN model, which outputs a predicted rating class (e.g., 1–5 stars) for each unobserved item, treated as a single-label classification target. To generate final Top-N recommendations, we further incorporate user-specific rating habits and item popularity to re-rank the classification outputs. The experimental results on public benchmark datasets indicate that our model substantially improves the prediction accuracy and recommendation quality compared with existing baselines. The proposed method offers a robust and interpretable approach to bridging textual review semantics, user behavior, and deep learning for rating-aware personalized recommendation.}, archive = {J_EAAI}, author = {Yanjun Xu and Chunqi Tian and Wei Wang and Lizhi Bai}, doi = {10.1016/j.engappai.2025.112196}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112196}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic analysis-based recommender system using sequential clustering and convolutional neural network}, volume = {161}, year = {2025}, }
• (2025). Ranking-oriented cross-modal hashing. EAAI, 161, 112195. (www)

  @article{ , abstract = {Deep cross-modal hashing has become a mainstream solution for multimedia retrieval. Most methods utilize pair-wise or multi-wise loss as proxies for ranking consistency, focusing primarily on local relational patterns. However, these schemes are not well aligned with global ranking metrics, which consider the relative order among all candidate items. This misalignment may push semantically similar items down the list. Additionally, such methods heavily rely on the selection of negative samples, which can cause training instability and require careful tuning. To address this problem, we propose a novel Ranking-Oriented Cross-Modal Hashing (ROCMH) framework, which performs global ranking optimization based on a vision-language model. Specifically, we design a differentiable ranking-oriented objective, called cross-modal ranking alignment loss. It smoothly simulates discrete ranking and naturally models ranking relations in candidate lists, thereby promoting more consistent cross-modal ranking. Meanwhile, considering the perceptual gap between visual and textual, we suggest a nanoparam training strategy with modality-aware prompts. This strategy greatly reduces the number of trainable parameters while ensuring that each modality provides unique signals for semantic information learning. Extensive experiments on three public datasets demonstrate that the method achieves superior retrieval performance and reduces the number of trainable parameters to less than 0.2% of the vision-language model. The source code is available: https://github.com/QinLab-WFU/ROCMH .}, archive = {J_EAAI}, author = {Yadong Huo and Qibing Qin and Wenfeng Zhang and Lei Huang}, doi = {10.1016/j.engappai.2025.112195}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112195}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ranking-oriented cross-modal hashing}, volume = {161}, year = {2025}, }
• (2025). Matching quality-guided model-free satellite pose estimation. EAAI, 161, 112194. (www)

  @article{ , abstract = {With the development of learning-based techniques and large-scale public datasets, satellite pose estimation has seen significant progress for the past several years. However, most of current methods still rely on a known three-dimensional (3D) model of the object for the pose estimation, limiting its generalization and wide application. To this end, we propose a model-free pose estimation method, which takes as input only a set of images. The proposed method consists of two stages, i.e. the reconstruction stage and pose estimation stage, of which the former reconstructs a 3D model from the input images and the pose is estimated via two-dimensional (2D)-3D feature matching by the latter. More importantly, a matching quality guidance strategy is introduced to further improve the robustness to in-plane rotation during feature matching. Additionally, since no known 3D model is assumed, our method generalizes well to novel objects without retraining. We provide evaluation results on datasets of several satellites with different structures, which demonstrate impressive performances without known models against current methods.}, archive = {J_EAAI}, author = {Zhaoshuai Qi and Yating Liu and Yanning Zhang}, doi = {10.1016/j.engappai.2025.112194}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112194}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Matching quality-guided model-free satellite pose estimation}, volume = {161}, year = {2025}, }
• (2025). Prebuilt spatiotemporal index: An exploration of efficient real-time data storage in intelligent transportation systems. EAAI, 161, 112193. (www)

  @article{ , abstract = {With the widespread deployment of Internet of Things (IoT) devices in Intelligent Transportation Systems (ITS), real-time spatiotemporal data has grown rapidly. Hybrid data-bases have emerged as the mainstream solution for managing such data. The LSM R*-tree, which integrates Log-Structured Merge-trees (LSM-trees) for time-series ingestion and R*-trees for spatial queries, is now a widely used spatiotemporal index. However, storage schemes based on LSM R*-tree structures involve real-time index construction, leading to significant overhead and write latency. To address these challenges, this paper proposes a prebuilt index-based data storage workflow that shifts index construction ahead of data writing. This approach allows the database to directly apply the prebuilt index at write time, thereby minimizing real-time construction costs and enhancing write performance. To support index prebuild, we propose a lightweight embedding scheme, Index2Vec, specifically designed for R*-tree structures. Based on this, we extend the Transformer architecture and develop the R*-tree Prediction Network for ITS (RTPN4ITS), which achieves efficient inference on resource-constrained edge devices. Experimental results show that the prebuilt R*-tree index improves query efficiency by up to 90% over time-index-only schemes and enhances write performance by nearly 50% compared to real-time indexing. The proposed RTPN4ITS model achieves robust accuracy across varying traffic densities, reaching 85% accuracy in dense conditions. Moreover, the Index2Vec embedding enhances the Transformer’s structural awareness. In summary, this paper proposes an efficient prebuilt indexing strategy and lightweight embedding-based model for real-time spatiotemporal data management in ITS.}, archive = {J_EAAI}, author = {Yiran Shao and Kangshuai Zhang and Yong Zhou and Zhenwu Chen and Yang Yang and Lei Peng}, doi = {10.1016/j.engappai.2025.112193}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112193}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prebuilt spatiotemporal index: An exploration of efficient real-time data storage in intelligent transportation systems}, volume = {161}, year = {2025}, }
• (2025). An error complementarity-based iterative learning approach via categorical boosting for student performance prediction. EAAI, 161, 112192. (www)

  @article{ , abstract = {Predicting student performance is important in achieving academic success and student development in many educational applications (e.g., academic early warning and early interventions). To accurately predict student performance, we propose an error complementarity-based iterative learning approach for student performance prediction. Our goal is to improve the prediction performance by iteratively reducing the prediction error. Specifically, in model construction, we first select the most important features to train the target estimator. Then, we obtain the errors between the target and predicted values. These errors are used to train the error estimator using the remaining features. Similarly, we iteratively train the model to learn from the errors until the desired conditions are met. In model prediction, we predict the testing sample using the target estimator to obtain the predicted value. Next, we predict the error for the next iteration using the corresponding error estimator. This process is repeated, and the final prediction is obtained by adding the predicted value and the error values. The extensive experiments from different educational datasets show that by using our error complementarity-based iterative learning approach, the proposed model outperforms the competing models for prediction accuracy. Furthermore, statistical testing conducted over 20-run experiments confirms the significant advantage of our proposed model. This suggests the iterative learning is effective for predicting student performance.}, archive = {J_EAAI}, author = {Zongwen Fan and Jin Gou and Cheng Wang}, doi = {10.1016/j.engappai.2025.112192}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112192}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An error complementarity-based iterative learning approach via categorical boosting for student performance prediction}, volume = {161}, year = {2025}, }
• (2025). Federated cross-device cluster dynamic time planning warping algorithm in the industrial internet of things. EAAI, 161, 112191. (www)

  @article{ , abstract = {Production lines and heterogeneous devices generate diverse time-series data in the Industrial Internet of Things (IIoT). This diversity necessitates frequent updates of local models when using federated learning (FL). FL has become widespread in the IIoT, these continual model updates can hinder timely information exchange among cross-industry devices. They constrain the system's ability to handle high concurrency in read and write operations. Therefore, we propose a federated cross-device cluster dynamic time planning warping (Fed-cDTW) tailored for the IIoT. We incorporate each device's historical behavior data by constructing a functional characteristic distance matrix. We design a symmetric time warping lower bound function to measure the similarity among multi-source time-series data. This approach enables an adaptive, dynamically organized industrial device cluster. Meanwhile, we employ the federated dynamic (FedD) dual-weight optimization within the device cluster to enhance the models' generalization and robustness. The experimental results evaluated the performance of the proposed method. We compare our method with baselines, Fed-cDTW improves classification accuracy by 0.8 percentage points (pp), 1.7 pp, 9.6 pp, 14.4 pp, and 49.5 pp. Convergence is accelerated by 0.04, 0.05, 0.35, 0.42, and 0.46. Total communication cost is reduced by 40 megabytes (MB), 56 MB, 84 MB, 120 MB, and 168 MB.}, archive = {J_EAAI}, author = {Yifan Zhao and Shibao Sun and Pengcheng Zhao and Yatong Wang and Jianfeng Liu}, doi = {10.1016/j.engappai.2025.112191}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112191}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated cross-device cluster dynamic time planning warping algorithm in the industrial internet of things}, volume = {161}, year = {2025}, }
• (2025). Machine learning-based prediction of compressive strength in sustainable self-compacting concrete. EAAI, 161, 112190. (www)

  @article{ , abstract = {The performance and durability of conventional concrete (CC) is significantly influenced by supplementary cementitious materials (SCMs) and recycled coarse aggregate (RCA). Thus, the intrusion of enrich SCMs with RCA in the cementitious matrix delivers utmost properties. This research focuses on the use SCMs and RCA in the self-compacting concrete (SCC) system and their sustainability in the development of concrete resources. Standard experimental methods for forecasting the compressive strength (CS) of SCC have constraints in terms of efficiency, time consumption, and cost. Thus, its prediction is crucial without the need for laborious experimental procedures. This work employs and evaluates a multiplicity of machine learning (ML) models to predict the CS of the cementitious matrix to tackle these issues. Thus, considered a more effective and cost-saving solution in comparison with traditional approaches. Therefore, ML approaches like, gene expression programming (GEP), decision trees (DT), and support vector regression (SVR) were employed. The performance of the model is evaluated by employing the coefficient of determination (R 2 ), statistics, and uncertainty analysis. Individual Conditional Expectation (ICE), and Partial Dependence Plot (PDP) are used to analyze the effect of parameters on strength. The findings suggest that GEP performs best achieving superior R 2 > 0.90 for the training, validation, and test data sets with a lower error. While, the uncertainty analysis shows that all modeled values lie below the threshold value. The ICE and PDP graphs confirms that cement, age, and water-cement ratio have highly relation to outcomes. The RCA replacement ratio is more important than the CA one, and SCMs play an essential role in the development of concrete compressive strength, although not as much as cement. In addition, SP depicts major contribution to SCC. Moreover, graphical user interface (GUI) is also developed to help users/researcher that will facilitate them to estimate the strength of SCC in practical applications.}, archive = {J_EAAI}, author = {Jingguo Gou and Athar Zaman and Furqan Farooq}, doi = {10.1016/j.engappai.2025.112190}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112190}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based prediction of compressive strength in sustainable self-compacting concrete}, volume = {161}, year = {2025}, }
• (2025). A robust fuzzy twin support vector machine with kernel-target alignment for binary classification. EAAI, 161, 112189. (www)

  @article{ , abstract = {Many algorithms similar to the twin version of the support vector machine and their variants have shown better results in the binary classification of nonlinear data points. But in the presence of outlier and noise, these algorithms exhibit low generalization efficiency. To alleviate this challenge, recently proposed, kernel-target alignment based fuzzy least square twin bounded support vector machine (KTA-FLSTBSVM) used fuzzy membership values and is solved using least squares. Inspired by this strategy, for further improvement, we propose a novel approach called decision support kernel-target alignment based fuzzy least square twin bounded support vector machine (DS-KFIFTBSVM). DS-KFIFTBSVM considers the kernelized fuzzy membership values with the regularized twin support vector machine and solves for linear and nonlinear data points using a functional iterative approach. In DS-KFIFTBSVM, the solution is obtained by solving a linearly convergent iterative scheme rather than solving quadratic programming problems. The proposed DS-KFIFTBSVM offers better generalization efficiency, which has been evaluated using both linear and Gaussian kernels, mostly on artificially developed and publicly accessible datasets with diverse dimensionalities. In terms of various performance evaluation metrics, including specificity, precision, false positive rate, rate of misclassification error, F_score, and geometric mean in linear and non-linear cases, DS-KFIFTBSVM outperforms various baseline approaches. It shows the highest accuracy in various datasets, including 98.1884 % for musk dataset (linear kernel) and 100 % for the glass dataset (Gaussian kernel). Further statistical analysis confirms its classification efficiency.}, archive = {J_EAAI}, author = {Deepak Gupta and Barenya Bikash Hazarika and Umesh Gupta and Witold Pedrycz}, doi = {10.1016/j.engappai.2025.112189}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112189}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust fuzzy twin support vector machine with kernel-target alignment for binary classification}, volume = {161}, year = {2025}, }
• (2025). Comprehensive assessment of pore-scale segmentation techniques for image-based multiphase flow characterization in porous media. EAAI, 161, 112188. (www)

  @article{ , abstract = {Evaluation of multiphase fluid flow behavior based on pore-scale imaging is significantly influenced by the accuracy of the segmentation techniques. This study sheds light on the influence of various traditional and Artificial Intelligence (AI) algorithms on the major static, dynamic, volumetric, and topological properties of the porous media. The traditional methods include the Gaussian Mixture Model (GMM), Multi-Otsu’s thresholding (MOT), and Random Walk (RW), and the AI-based techniques include Trainable Weka (TW), as well as seven deep convolutional autoencoder architectures. The assessment was carried out by comparing the calculated phase fraction, contact angle, capillary pressure, effective permeability, and topology for multiphase flow images of Bentheimer sandstone. The results revealed that combining Residual Networks (ResNet) and UNet structures (UResNet) marginally performs better based on the statistical metrics. Compared to ground-truth watershed images, applying other approaches except TW yielded acceptable static and volumetric parameters. The main discrepancies were in contact angles, capillary pressure, and phase topology. These discrepancies also influenced calculated effective permeabilities, yielding unrealistic trends. UResNet provided the best accuracy, considering the watershed the ground truth. MOT and RW showed similar performance, while GMM performed inconsistently throughout the analysis and TW produced acceptable results for certain fractional flows. The segmentation approaches affected topological features most, whereas properties such as phase fractions were less affected. The inconsistent behavior and effectiveness of used algorithms indicate that the analysis of multiphase images necessitates a meticulous choice of segmentation techniques, emphasizing AI-based algorithms that ensure consistency across all evaluations.}, archive = {J_EAAI}, author = {Javad Siavashi and Mohammad Sharifi}, doi = {10.1016/j.engappai.2025.112188}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112188}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive assessment of pore-scale segmentation techniques for image-based multiphase flow characterization in porous media}, volume = {161}, year = {2025}, }
• (2025). Land use and land cover change detection from multisource satellite imagery – A hybrid convolutional neural network approach. EAAI, 161, 112187. (www)

  @article{ , abstract = {Remote sensing and satellite imagery analysis enable us to observe and understand environmental changes over time. However, such an endeavor can be complex and time-consuming. To overcome this challenge, we propose a framework for leveraging Google Earth Engine (GEE) to analyze Earth's surface changes over time. Our framework starts by assessing land cover changes using satellite imagery, evaluating the effectiveness of vegetation indices in classification, and generating accurate land cover maps for a specific region. We examine the validity of this framework on the area near Rio Do Sul, located in the Santa Catarina district of Brazil, to classify and recognize the alteration in land use and land cover (LULC) from 2019 to 2023. In this effort, we develop a pre-trained Convolutional Neural Network (CNN) for feature extraction from the Landsat 8, Sentinel-1, and Sentinel-2 images and then link those into the random forest (RF) algorithm for spatial morphology and temporal change logic to map the long-term annual time series and detect changes in the region at hand. The novelty of this study arises from developing a CNN-RF hybrid model and implementing a data preprocessing pipeline, which includes progressive techniques for cloud masking and radiometric calibration, to ensure higher accuracy and reliability in land use and land cover classification. This CNN-RF hybrid model successfully traced the increasing and decreasing rate in the built-up area, water bodies, forest, fallow land, plantations, and grassland with an overall accuracy of 96.39 % and 94.15 % for 2019 and 2023, respectively.}, archive = {J_EAAI}, author = {M.S. Babitha and A. Diana Andrushia and N. Anand and M.Z. Naser and Y. Pari}, doi = {10.1016/j.engappai.2025.112187}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112187}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Land use and land cover change detection from multisource satellite imagery – A hybrid convolutional neural network approach}, volume = {161}, year = {2025}, }
• (2025). Safety monitoring digital twin-based centralized model consolidation mechanism using dynamic node selection for multi-worker safety prediction. EAAI, 161, 112186. (www)

  @article{ , abstract = {The construction industry remains one of the most hazardous sectors, requiring innovative solutions to safeguard workers, especially in complex, dynamic outdoor environments. Digital Twin (DT) technology offers promising capabilities for real-time safety monitoring through virtual replicas of physical systems. However, existing DT frameworks rarely integrate comprehensive safety monitoring via a centralized model consolidation mechanism, such as Federated Learning (FL), which is explicitly tailored for multi-worker scenarios. Addressing this challenge, this paper proposes a FL-based Safety Monitoring Digital Twin (SMDT) framework designed to enhance multi-worker safety in resource-constrained settings. This enables real-time safety monitoring and control by representing on-site workers as virtual objects within a synchronized DT environment. A dynamic node selection mechanism based on client performance is employed to optimize global model convergence in FL. To validate the proposed approach, an edge computing-based experimental testbed using actual Raspberry Pi devices was implemented, using real-world construction safety data including worker status, weather conditions, and building structural parameters. Experimental results demonstrate the effectiveness of the proposed framework in significantly improving safety predictions and real-time monitoring efficiency. This research establishes a foundational work towards safer construction sites through intelligent, synchronized safety monitoring systems.}, archive = {J_EAAI}, author = {Sa Jim Soe Moe and Atif Rizwan and Anam Nawaz Khan and Rongxu Xu and Do Hyeun Kim}, doi = {10.1016/j.engappai.2025.112186}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112186}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Safety monitoring digital twin-based centralized model consolidation mechanism using dynamic node selection for multi-worker safety prediction}, volume = {161}, year = {2025}, }
• (2025). Universal entity linking. EAAI, 161, 112185. (www)

  @article{ , abstract = {Entity linking is a process of connecting mentions of entities in a document to corresponding entries in a knowledge base. Traditional models for entity linking often require specific fine-tuning to work with knowledge bases other than those they were originally pretrained on, which limits their flexibility and scalability. Building on the concept of entity profile generation, we propose a novel approach that enables entity linking across various knowledge bases without the need for such fine-tuning. Our pipeline leverages a fine-tuned Large Language Model, a generic embedding model, and a vector store to achieve high precision on the TweekiGold and Reuters-128 datasets. Additionally, it demonstrates strong retrieval rates across the TweekiGold , Reuters-128 , and ISTEX-1000 Wikidata entity linking datasets. We also illustrate the applicability of our method to other knowledge bases, using the Agrovoc knowledge base as an example. This solution offers a more versatile and scalable approach to entity linking.}, archive = {J_EAAI}, author = {Adam Aron Rynkiewicz and Raul Palma and Piotr Formanowicz}, doi = {10.1016/j.engappai.2025.112185}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112185}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Universal entity linking}, volume = {161}, year = {2025}, }
• (2025). A search method for fractured-vuggy reservoir inter-well connectivity path based on multi-modal multi-agent. EAAI, 161, 112184. (www)

  @article{ , abstract = {The complex geological structure of carbonate reservoirs and the intricate fracture-vuggy configurations obscure inter-well connectivity, making its evaluation challenging. Conventional studies primarily rely on seismic static data to delineate fracture-vuggy reservoirs, but the limited recognition accuracy hampers the precise characterization of inter-well connectivity and the spatial configuration of fractures and vugs. To address this, this study constructs a 3D (Three-Dimensional) search environment and use multi-modal static and dynamic data and proposes a multi-agent connected channel search model based on deep reinforcement learning. The model treats multiphase fluid as an agent and incorporates Swin Transformer (Shift Window Transformer) to extract large-scale fracture features from seismic data, providing global prior information for path search. A Graph Attention Network is established based on dynamic response relationships to extract spatial geological features, while a multi-head self-attention mechanism captures real-time fluid interactions in various directions. The model fuses multi-modal features, including seismic attributes and production data, to generate decisions and automatically search for inter-well connectivity channels. Experiments were conducted using the WE1 and WE5 well groups from the fault-controlled karst reservoirs in the Tahe oilfield, with results compared against tracer tests. The findings demonstrate that the proposed model's automatic search paths closely align with seismic data and tracer test results, effectively capturing the spatial distribution of fractures and vugs across different scales. This validates the model's effectiveness in evaluating inter-well connectivity in complex carbonate reservoirs.}, archive = {J_EAAI}, author = {Wenbin Jiang and Dongmei Zhang and Hong Cao and Xiaofeng Wang}, doi = {10.1016/j.engappai.2025.112184}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112184}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A search method for fractured-vuggy reservoir inter-well connectivity path based on multi-modal multi-agent}, volume = {161}, year = {2025}, }
• (2025). Continuous spatio temporal prompts for visual tracking. EAAI, 161, 112183. (www)

  @article{ , abstract = {Currently, visual single-object tracking methods utilize online template updates to combine temporal information. However, these methods rely on confidence scores to evaluate the reliability of the current template, which may result in a template not being updated for an extended period. Moreover, advanced trackers select bounding boxes based solely on the similarity between the template and the search area, which can lead to tracking drift when encountering deformable or similar targets. To alleviate these limitations, we propose a Spatio Temporal Prompt Tracker (STPTrack), which utilizes the prior information about small changes of object state between successive frames. Different from previous tracking methods that mainly rely on templates and similarity scores, STPTrack transfers the object position and shape information of the previous frame to the current frame as continuous spatio temporal prompt for the first time, and realizes the efficient fusion of spatio temporal information through the prompt encoder and the fusion decoder module. Specifically, it encodes the bounding box coordinates or mask information of the previous frame and the response points of the current frame as prompt features, and then combines prompt tokens with search tokens through the fusion decoder to provide the potential location of the object for the search feature map. Our STPTrack sets a new state-of-the-art performance on six tracking benchmark datasets.}, archive = {J_EAAI}, author = {Meng Sun and Xiaotao Liu and Yifan Li and Hongyu Wang and Dian Yuan and Jing Liu}, doi = {10.1016/j.engappai.2025.112183}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112183}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continuous spatio temporal prompts for visual tracking}, volume = {161}, year = {2025}, }
• (2025). Length-aware center loss for sequence to sequence thai scene text recognition. EAAI, 161, 112182. (www)

  @article{ , abstract = {Thai scene text recognition is a challenging task because Thai can be written in both horizontal and vertical directions, allowing characters to be stacked vertically. To address this issue, our previous work combined vertically stacked characters to create new characters. However, this strategy introduced many similar characters. In this paper, we further investigate this problem and propose the Length-aware Center Loss (LC) for Thai scene text recognition. The original center loss was designed for single object recognition tasks. When applied to multi-label tasks like text recognition, center loss is only effective when the lengths of the labels and prediction results are consistent. This can lead to an extreme case where all images receive incorrect predicted text lengths to minimize loss, severely interfering with the recognition process. Therefore, we propose the Length-aware Center Loss for text recognition. We also design the Length Supervision Module (LSM) and the Feature Clustering Module (FCM) to work alongside the LC loss. LSM predicts text length to provide additional supervision signals, while FCM aims to improve recognition performance by minimizing the distance between the features of corresponding class centers. Since there is no publicly available Thai scene text dataset, we have collected a new dataset containing more than 170,000 samples. Extensive experiments conducted on this dataset show that our method achieves superior performance in both string-level and character-level accuracy compared to other methods.}, archive = {J_EAAI}, author = {Hongjian Zhan and Chun Li and Bing Yin and Yue Lu}, doi = {10.1016/j.engappai.2025.112182}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112182}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Length-aware center loss for sequence to sequence thai scene text recognition}, volume = {161}, year = {2025}, }
• (2025). Diffusion-based vision-language model for zero-shot anomaly detection in medical images. EAAI, 161, 112181. (www)

  @article{ , abstract = {With the rapid advancement of diagnostic technology, the ability to detect pathological areas such as tumors and polyps has significantly improved. This progress provides medical imaging specialists with more precise visual information to support anomaly identification, diagnosis, treatment planning, and patient monitoring. However, existing unsupervised and semi-supervised anomaly detection methods struggle with data privacy constraints, limited annotated medical datasets, and challenges in generalization. Zero-Shot Anomaly Detection (ZSAD), which enables the detection of unseen categories without requiring class-specific training, has emerged as a promising solution by leveraging the vision-language alignment capabilities of Vision-Language Models (VLMs), such as Contrastive Language-Image Pretraining (CLIP). Despite recent progress, ZSAD remains hindered by high noise levels, sparse targets, and poor adaptability in complex medical imaging scenarios. To address these issues, we propose a novel framework: DiffusionCLIP, a diffusion-based VLM for zero-shot anomaly detection in two-dimensional medical images. Specifically, DiffusionCLIP integrates diffusion models into the VLM to progressively denoise multi-level features extracted from the CLIP visual encoder, enhancing feature robustness and discriminability. A multi-level feature fusion strategy is designed to aggregate multi-scale representations from different depths of the visual encoder, ensuring complementary semantic alignment across layers. In addition, a dynamically modulated weight loss function is introduced to adaptively balance the learning of hard and easy samples, further improving model generalization. Extensive experiments on multiple benchmark medical imaging datasets, demonstrate that the proposed method significantly outperforms existing zero-shot anomaly detection approaches in terms of accuracy, robustness, and generalization.}, archive = {J_EAAI}, author = {Yanhui Chen and Hongkang Tao and Zan Yang and Yunkang Cao and Chen Jiang and Longhua Hu and Pengwen Xiong and Haobo Qiu}, doi = {10.1016/j.engappai.2025.112181}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112181}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diffusion-based vision-language model for zero-shot anomaly detection in medical images}, volume = {161}, year = {2025}, }
• (2025). Surface roughness prediction in turning processes for grey cast iron: A hybrid machine learning approach integrating infrared thermography. EAAI, 161, 112180. (www)

  @article{ , abstract = {Workpiece surface quality is a critical control parameter in machining processes, influencing functional performance, dimensional precision, and wear resistance. However, accurately predicting surface roughness is complex, often limited by the computational demands of traditional high-precision methods and the reliance of existing models solely on cutting parameters, hindering real-time monitoring. This research introduces a hybrid Artificial Neural Network (ANN) methodology specifically developed for predicting surface roughness of grey cast iron GG-25 workpieces machined in turning processes, a material previously unstudied in this context. The methodology integrates real-time infrared thermal measurements from multiple defined regions of interest (ROIs) within the tool-workpiece contact zone, along with cutting parameters. Experimental results demonstrated that feed rate (f) is the most significant cutting parameter (effect = 0.43) affecting surface quality, followed by its combination with cutting speed (V c ) (effect = −0.25) and cutting speed (effect = 0.18). Correlation and non-linear regression analyses revealed complex, often exponential relationships between temperature and surface roughness, showing temperature an upward trend as machining progressed. The developed ANN achieves a correlation coefficient (R) value of 0.99 both when predicting the roughness arithmetic mean deviation (Ra) parameter in training conditions and when using data from experiments not used in training (validations data). Moreover, the model reaches a correlation coefficient value of 0.85 (test data) under cutting conditions different from those used in experiments, demonstrating robustness, significantly outperforming Support Vector Regression (SVR). This model represents a highly potential tool for real-time online inspection.}, archive = {J_EAAI}, author = {Sergio Aguado and Marcos Pueo and Raquel Acero and Ana Cristina Majarena and Jorge Santolaria}, doi = {10.1016/j.engappai.2025.112180}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112180}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Surface roughness prediction in turning processes for grey cast iron: A hybrid machine learning approach integrating infrared thermography}, volume = {161}, year = {2025}, }
• (2025). Legendre multiwavelet-based feature attention guidance lightweight network for accurate steel surface defect classification. EAAI, 161, 112179. (www)

  @article{ , abstract = {Steel surface defects can significantly affect the quality and appearance of industrial products such as aerospace, construction application fields, and so on. Due to the multi-scale morphology, low contrast, and random positions of the defects, achieving a favorable balance between detection accuracy and speed remains challenging in practical applications. To address these challenges, this paper combines Legendre multiwavelet (LW) with feature attention guidance (FAG) mechanism to devise a novel high-accuracy lightweight network (LWFAG-LNet) for surface defect classification. More precisely, LW bases with rich regularities are first utilized to match complex geometric characteristics across multi-wavelet and multi-scale resolution levels. Subsequently, the FAG module effectively fuses shallow high-resolution detailed features with deep low-resolution contextual features, significantly reducing the depth of the convolutional neural network (CNN). To the third step, a lightweight CNN module with four convolutional blocks is designed to further extract deep features while preserving the most valuable defect information. The proposed model achieves the highest recognition accuracy with a simple structure and a compact parameter configuration. Extensive experiments are conducted on the Northeastern University-Classification (NEU-CLS), Xsteel surface defect dataset (X-SDD), and Kungliga Tekniska Högskolan Royal Institute of Technology Textures under varying Illumination, Pose and Scale (KTH-TIPS) dataset to verify the model's effectiveness and generalization capability. The results reach classification accuracies of 99.80 %, 98.49 %, and 99.06 %, outperforming existing models by about 1.45 %, 3.45 %, and 2.04 %, respectively. In summary, the proposed LWFAG module can be flexibly applied to various lightweight network frameworks, demonstrating great potential for real-time surface defect detection.}, archive = {J_EAAI}, author = {Xiaoyang Zheng and Weishuo Liu and Yan Huang}, doi = {10.1016/j.engappai.2025.112179}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112179}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Legendre multiwavelet-based feature attention guidance lightweight network for accurate steel surface defect classification}, volume = {161}, year = {2025}, }
• (2025). Detecting object hue bin in the CIECAM02 uniform color space using a multi-layer perceptron for optimal spectrum generation. EAAI, 161, 112178. (www)

  @article{ , abstract = {In this paper, we present an innovative enhancement to an existing multi-channel light-emitting diode luminaire by integrating artificial intelligence and image analysis capabilities. The system captures an image of the object to be illuminated using a custom-developed Android application. The dominant color in the image is first identified, and then a multi-layer perceptron neural network classifies it into one of 16 hue bins within a uniform color appearance model. This classification process enables the luminaire to select a predefined spectrum designed to enhance the visual appeal of the illuminated object. To ensure robust hue bin classification, we developed a rich dataset with 7920 samples, collected using eleven smartphones under ten distinct light sources. Using this diverse dataset, thirteen multi-layer perceptron neural networks with varying input features were trained and compared, achieving hue classification accuracies, precision, and recall all ranging from 95 % to 99 % on the test set. One of the trained multi-layer perceptron networks was integrated into the Android application, with an inference time of 90 μs on a mid-range smartphone, demonstrating the practical application of artificial intelligence for on-demand hue analysis and lighting optimization. This advancement not only enhances the luminaire's functionality but also enriches the user experience by providing intelligent, on-demand lighting solutions tailored to the object's dominant color.}, archive = {J_EAAI}, author = {Esmat Kishani Farahani and Kaveh Ahmadian Tazehmahaleh}, doi = {10.1016/j.engappai.2025.112178}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112178}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detecting object hue bin in the CIECAM02 uniform color space using a multi-layer perceptron for optimal spectrum generation}, volume = {161}, year = {2025}, }
• (2025). Generalized zero-shot fault diagnosis method for chillers based on cross-modal information compression. EAAI, 161, 112177. (www)

  @article{ , abstract = {In practical chiller fault diagnosis applications, target fault training data is often inaccessible, which poses significant challenges to the effectiveness of data-driven diagnostic approaches. Generalized zero-shot fault diagnosis (GZSFD) aims to detect and classify all fault types without relying on fault samples from all categories during training. GZSFD models are trained exclusively on data from seen faults—those with available historical data—making them prone to bias toward these seen classes during inference. However, existing methods fail to exploit the fine-grained semantics of fault attributes, resulting in suboptimal performance when addressing feature bias toward seen classes. In this article, an innovative GZSFD framework for chillers based on cross-modal information compression is proposed to overcome this difficulty. This method constructs a cross-modal feature fusion attention (CmFFA) and information compression module, and integrates them into a variational autoencoder with generative adversarial network (VAEGAN), and establishes a CmFFA-VAEGAN network to synthesize high-quality unseen fault samples. Specifically, the CmFFA module effectively aligns local regions of the virtual samples with key words in the textual attributes, enabling the model to attend to fine-grained semantic information and robustly fuse features from different modalities. This facilitates the generation of virtual samples that more accurately reflect the characteristics of the target modality. In addition, an information bottleneck (IB) layer is introduced at the output of the generator to compress redundant information within the fused features, retaining only the key information most relevant to the data augmentation task. This design enhances cross-modal consistency between the synthesized samples and their corresponding attributes, while alleviating the feature shift of unseen class samples toward seen categories. Extensive experiments are designed and executed on the chiller dataset. Experimental results demonstrate that the proposed framework effectively mitigates the bias of synthesized unseen fault samples toward seen categories, leading to a significant improvement in diagnostic accuracy.}, archive = {J_EAAI}, author = {Kexin Jiang and Xuejin Gao and Huayun Han and Huihui Gao and Yongsheng Qi and Xi Zhang and Liang Zhao}, doi = {10.1016/j.engappai.2025.112177}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112177}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generalized zero-shot fault diagnosis method for chillers based on cross-modal information compression}, volume = {161}, year = {2025}, }
• (2025). Enhancing rotation you only look once version 8 for accurate detection of arbitrarily-oriented and multi-scale construction material in complex environment. EAAI, 161, 112176. (www)

  @article{ , abstract = {Construction material detection is integral to effective material management. While deep learning-based methods have advanced automatic detection, challenges such as arbitrarily oriented objects, similar features, large object scale variations, and noise interference still limit accuracy. This study proposes an enhanced detection method based on improved rotation you only look once version 8, incorporating three key innovations. First, the large selective kernel network employs a spatial selection mechanism to dynamically adjust the network's receptive field, capturing key features and contextual information of various materials in complex construction scenarios. This enhances object detection performance in feature-similar environments. Second, the poly kernel inception network combines non-dilated multi-scale convolutions to extract dense texture features from differently sized objects, while using contextual anchor attention to capture long-range information for small objects. These components work together to improve multi-scale object detection. Third, the rectangular self-calibration module uses horizontal and vertical pooling to model rectangular attention regions, capturing axial global context. Its shape self-calibration function adjusts these regions to better fit arbitrarily oriented construction materials, enhancing focus on objects while suppressing noise interference. Experimental results show a mean average precision of 0.871–2.7 % higher than the baseline model and surpassing other state-of-the-art methods—while maintaining real-time detection speed at 28.3 frames per second. The proposed method improves material detection accuracy in complex construction environments, enabling refined material management on-site. This supports material entry and exit tracking, on-site usage monitoring, inventory management, and procurement planning, while also strengthening lean control of construction progress and costs.}, archive = {J_EAAI}, author = {Yujie Lu and Yuanjun Nong and Dayu Zhu and Shuo Wang}, doi = {10.1016/j.engappai.2025.112176}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112176}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing rotation you only look once version 8 for accurate detection of arbitrarily-oriented and multi-scale construction material in complex environment}, volume = {161}, year = {2025}, }
• (2025). Long-term forecasting of monthly reservoir inflow using deep and machine-learning-based algorithms. EAAI, 161, 112175. (www)

  @article{ , abstract = {This study provides a method for dynamic flow forecasting to detect how to move between periods with high and low-flow peaks. In other words, the model continues by gaining a deep insight into the long-term stable changes of the time series. Four-time horizons (5, 10, 15 and 20 years) were chosen to forecast the streamflow. These time horizons represent the ability of this method to detect the continuation of its movement in the short, medium, and long term. In this study, a dynamic and recurrent cycle is used to forecast river streamflow based on long-short-term memory (LSTM) and artificial neural network (ANN) models. The 12-month time delay of precipitation and streamflow of the Zayanderud river from 1970 to 2019 was used as input variables for training and validation of the models. In each stage of the forecasting process, 1 year (12 months) is forecasted and used as input for the next stage. The results show that the ANN model performs better in shorter time horizons and the LSTM model in longer time horizons, respectively. The best performance of the ANN model occurs in the 10-year forecast (Mean Absolute Percentage Error (MAPE) = 44.3, Nash–Sutcliffe efficiency coefficient (NSCE) = 0.58, Root Mean Square Error (RMSE) = 43.75) and the best performance of the LSTM model occurs in the 15-year forecast (MAPE = 48.9, NSCE = 0.54, RMSE = 47.12). Finally, the selected period is used to forecast the streamflow in future periods when observational data are not available.}, archive = {J_EAAI}, author = {Bahram Ghenaati and Mohammad Reza Nikoo and Mohammad G. Zamani}, doi = {10.1016/j.engappai.2025.112175}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112175}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long-term forecasting of monthly reservoir inflow using deep and machine-learning-based algorithms}, volume = {161}, year = {2025}, }
• (2025). Characteristics detection of surface agglomerated abrasive on electroplated diamond wire: A measurement method based on semantic segmentation. EAAI, 161, 112173. (www)

  @article{ , abstract = {Electroplated Diamond Wire (EDW) serves as the primary tool for cutting semiconductor materials. Abrasive agglomeration on its surface induces sudden fluctuations in sawing force, exacerbates material surface damage, and shortens tool lifespan. However, existing testing methods primarily focus on abrasive counts while generally lacking effective means to quantify key characteristics of agglomerated abrasives (e.g., density, maximum size, and number of abrasives contained), leading to substantial limitations in the EDW quality assessment system. To address this critical gap, this study proposes a U-shape Network (UNet)-based semantic segmentation model for EDW surface abrasive grain segmentation, termed Electroplated Diamond Wire U-shape Network (EDW-UNet). By integrating Atrous Spatial Pyramid Pooling (ASPP) and Mamba-like Linear Attention (MLLA), EDW-UNet enhances the detection of multi-scale targets and the accuracy of complex boundary segmentation, enabling precise segmentation of both single and agglomerated abrasives on EDW surfaces. Experimental results demonstrate that EDW-UNet achieves a mean Intersection over Union of 0.917 and a mean Pixel Accuracy of 0.961 on the test set, outperforming the baseline UNet and other mainstream models, with an inference speed of up to 12.82 frames per second. For the first time, a feature extraction algorithm developed from the segmentation results enables measurement of key parameters, including the density, maximum size, and number of abrasives contained in agglomerates. This study not only addresses the technical gap in quantitative detection of agglomerated abrasive features but also provides a novel high-precision, quantitative standard for optimizing EDW manufacturing processes and assessing product quality, thereby enhancing the cutting efficiency and yield of semiconductor materials. The code is available at: https://github.com/huangwenbin123/EDW-UNet .}, archive = {J_EAAI}, author = {Wenbin Huang and Yufei Gao and Shengtan Hu and Dameng Cheng}, doi = {10.1016/j.engappai.2025.112173}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112173}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Characteristics detection of surface agglomerated abrasive on electroplated diamond wire: A measurement method based on semantic segmentation}, volume = {161}, year = {2025}, }
• (2025). Risk-response coupling in underground structures under liquefiable soil conditions: A causality-informed dynamic bayesian network integrated framework. EAAI, 161, 112171. (www)

  @article{ , abstract = {Underground structures in liquefiable soils face complex seismic risks that can trigger cascading failures. This study proposes a Granger causality-informed Dynamic Bayesian network (G-DBN) framework to capture the temporal propagation of seismic risk in such systems. Firstly, a system risk assessment model integrates multiple performance indicators through Cloud Model (CM) to quantify overall risk levels, considering uncertainties associated with soil liquefaction and structural responses. Subsequently, a structural dynamic risk inference model is established using Dynamic Bayesian network (DBN), combining Granger Causality (GC) analysis with engineering-informed relationships to define the network structure. The input features include key structural state variables such as tunnel cross-section convergence ( T 4 ), tunnel uplift displacement ( T 6 ), station uplift displacement ( S 5 ), and inter-story drift angle ( S 6 ), and the aggregated structural risk indicator serves as the target variable. This framework enables the temporal propagation of risk across interconnected structural nodes, and elucidates the mechanisms by which liquefiable soil deformations and structural responses interact within the soil-structure system. Results showed that the risk characteristic value (Expectation , E x ) decreased from 29.12 % (percentage) to 5.21 % as the Peak Ground Acceleration (PGA, expressed in units of gravitational acceleration g) increased from 0.1 g to 0.7 g. The proposed G-DBN model demonstrates robust predictive capabilities, achieving coefficient of determination ( R 2 ) values exceeding 0.95 across multiple seismic intensity conditions. Additionally, tunnel cross-section convergence ( T 4 ) was identified as the most critical factor affecting risk propagation in the coupled underground systems. By integrating holistic risk quantification with dynamic propagation analysis, this study offers a robust tool for understanding dynamic risk evolution and supports decision-making for seismic resilience of underground infrastructure in liquefiable soils.}, archive = {J_EAAI}, author = {Heqi Kong and Xiaohua Bao and Jun Shen and Xiangcou Zheng and Xiangsheng Chen}, doi = {10.1016/j.engappai.2025.112171}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112171}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk-response coupling in underground structures under liquefiable soil conditions: A causality-informed dynamic bayesian network integrated framework}, volume = {161}, year = {2025}, }
• (2025). Neural operator-enabled forward and inverse modeling of laser-induced surface acoustic waves and applications in nondestructive evaluation. EAAI, 161, 112170. (www)

  @article{ , abstract = {Laser-induced surface acoustic wave (SAW)-driven nondestructive evaluation offers high-resolution, non-contact characterization of subsurface microstructures. However, its practical application is often limited by the high computational costs associated with traditional numerical simulation methods. Recently, machine learning has emerged as an attractive alternative to accelerate these simulations. In this paper, we develop a neural operator-enabled framework for both forward and inverse modeling of laser-induced SAW propagation. A general dataset with randomly generated subsurface structures is used to evaluate and quantify the model's performance in both wave propagation and subsurface inversion problems. Three potential applications are then investigated: subsurface crack localization, multilayer structure characterization and polycrystalline grain imaging. The results demonstrate that the neural operator-enabled model achieves satisfactory accuracy even in the presence of noise and source waveform variations, underscoring its potential as an efficient and accurate surrogate model for practical nondestructive evaluation using laser-induced SAWs.}, archive = {J_EAAI}, author = {Zaiwei Liu and Zhongqing Su}, doi = {10.1016/j.engappai.2025.112170}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112170}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural operator-enabled forward and inverse modeling of laser-induced surface acoustic waves and applications in nondestructive evaluation}, volume = {161}, year = {2025}, }
• (2025). A multi-objective double Q-learning-based hyper-heuristic algorithm for aluminum production and transportation integrated scheduling problem. EAAI, 161, 112169. (www)

  @article{ , abstract = {The cooperative process of electrolytic aluminum production and aluminum liquid distribution is a typical distributed integrated scheduling problem. The aluminum production and transportation integrated scheduling problem (APTISP) is a strong coupling problem that is studied with the optimal objectives of total completion time and total energy consumption. A mathematical model is constructed for the APTISP. A multi-objective double Q-learning-based hyper-heuristic (MDQHH) algorithm is designed to solve the APTISP. In the MDQHH algorithm, a heuristic rule-based initialization operation is designed to generate the initial population of the APTISP. Nine problem-specific low-level perturbation heuristics are constructed to explore the APTISP solution space. A dynamically adjustable ε -greedy strategy is designed to select the low-level perturbation heuristics, and ε gradually decreases as the learning process progresses. Two Q-tables are utilized alternately to select actions and update the Q-values based on the current solution in each iteration. The experimental results show that the MDQHH algorithm outperforms several state-of-the-art comparison algorithms in addressing the performance of APTISP.}, archive = {J_EAAI}, author = {Fuqing Zhao and Ting Yang and Tianpeng Xu and Jianlin Zhang}, doi = {10.1016/j.engappai.2025.112169}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112169}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-objective double Q-learning-based hyper-heuristic algorithm for aluminum production and transportation integrated scheduling problem}, volume = {161}, year = {2025}, }
• (2025). Scalable multi-agent reinforcement learning for factory-wide dynamic scheduling in semiconductor manufacturing. EAAI, 161, 112168. (www)

  @article{ , abstract = {Real-time dynamic scheduling in modern manufacturing is highly complex due to frequent disturbances and intricate operational constraints. While reinforcement learning (RL) has shown promise, existing approaches often rely on extensive dispatching rules and struggle to scale to factory-wide settings. This study introduces a scalable multi-agent RL (MARL) framework with a leader–follower structure, enabling decentralized agents to handle sub-problems while maintaining global coordination through abstract goals. To further enhance robustness, a limited rule-based conversion algorithm is proposed to mitigate performance degradation from poor agent decisions. Our experimental results demonstrate that the proposed model outperforms the state-of-the-art deep RL-based scheduling methods in various aspects. Additionally, the proposed model provides the most robust scheduling performance to demand changes. Overall, the proposed MARL-based scheduling model presents a promising solution to the real-time scheduling problem, with potential applications in various manufacturing industries.}, archive = {J_EAAI}, author = {Jaeyeon Jang and Diego Klabjan and Han Liu and Nital S. Patel and Xiuqi Li and Balakrishnan Ananthanarayanan and Husam Dauod and Tzung-Han Juang}, doi = {10.1016/j.engappai.2025.112168}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112168}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scalable multi-agent reinforcement learning for factory-wide dynamic scheduling in semiconductor manufacturing}, volume = {161}, year = {2025}, }
• (2025). Research on global path planning algorithm based on indoor map partition preprocessing. EAAI, 161, 112167. (www)

  @article{ , abstract = {Mobile robots utilize Simultaneous Localization and Mapping (SLAM) technology to generate environmental maps and determine their locations within these environments. Subsequently, they employ path planning algorithms to complete navigation tasks. Although existing path planning algorithms are relatively mature, they still exhibit inefficiencies in complex indoor environments. To address this issue, this paper introduces an Indoor Map Partitioning Preprocessing (IMPP) algorithm, which identifies and segments irregularly shaped, complex rooms to accelerate the path planning process. The method initially utilizes the Robot Operating System (ROS) to construct an indoor map dataset and subsequently applies an image segmentation model to identify and enclose rooms. By combining image processing techniques with path planning algorithms, this method can obtain room index information and successfully exclude irrelevant areas from the path planning process. Ultimately, the IMPP algorithm is integrated with a variety of global path planning algorithms. Experimental results demonstrate that in complex indoor environments, this method significantly surpasses existing partitioning methods in terms of room recognition accuracy. Moreover, it decreases the number of expansion points in global path planning algorithms, significantly enhancing processing speed and efficiency.}, archive = {J_EAAI}, author = {Jifan Yang and Xiaoling Li and Xiaoyang Liu and Xunding Pan and Lei Wang}, doi = {10.1016/j.engappai.2025.112167}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112167}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on global path planning algorithm based on indoor map partition preprocessing}, volume = {161}, year = {2025}, }
• (2025). An ensemble transfer learning strategy for short-term electricity load forecasting of data-sparse buildings. EAAI, 161, 112166. (www)

  @article{ , abstract = {The “data silos” caused by the lack of available historical or shared information is an important reason affecting forecasting performance of building’s short-term power demand, thus limits the flexibility of demand response of building group. Knowledge’s transfer learning (TL) from similar domains has been proved an efficient tool to enhance target prediction’s performance. To improve data-sparse building’s energy forecasting accuracy and robustness simultaneously, an ensemble strategy based on different TL models is proposed in this study. Firstly, a two-stage similar data selection method is proposed for source domain construction. The multi-kernel maximum mean discrepancy (MK-MMD) is used for selecting similar buildings to the target one; the nearest neighbor search (NNS) method is adopted for picking the most helpful data from the similar buildings to form the dataset of source domain. To determine the optimal number of source domains for transfer learning, an objective function is also formulated. Then, two effective TL models, i.e., instance-based TL (IBTL) and model-based TL (MBTL) are both built as sub models. On this basis, a parallel ensemble framework is designed with model weights adjusted by multiple linear regression. For case study, a public data set containing 1,449 buildings is treated as candidate source domain set for knowledge transfer, and two educational buildings with sparse historical data are treated as target buildings for hourly electricity load forecasting. Forecasting results show that compared with the previous reported TL models, the proposed ensemble strategy always achieves the best performance in two cases and the prediction accuracies are improved by 41.8% and 29.3% (MAPE) respectively. The impact of source domain number and target data amount on prediction performance are also analyzed and discussed in details.}, archive = {J_EAAI}, author = {Xingqiao Liu and Borui Wei and Wenping Xue and Xiaoying Li and Kangji Li}, doi = {10.1016/j.engappai.2025.112166}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112166}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble transfer learning strategy for short-term electricity load forecasting of data-sparse buildings}, volume = {161}, year = {2025}, }
• (2025). Intelligent multispectral image recognition by using credibility-based square root cubature kalman filters and broad particle swarm learning. EAAI, 161, 112165. (www)

  @article{ , abstract = {This paper proposes intelligent multispectral image recognition by using credibility-based Square Root Cubature Kalman Filters (SRCKF) and broad particle swarm learning to address the issue of low recognition accuracy in multispectral target detection using Kalman filtering. This method initially enhances the estimation accuracy of SRCKF by integrating credibility theory. Secondly, the regression parameters of a Broad Learning System (BLS) are tuned with Particle Swarm Optimization (PSO), forming a PSO-BLS network that offers fast, lightweight feature learning. The PSO-BLS acts as an regressor that compensates residual errors produced by the credibility-based SRCKF, thereby refining the overall filtering precision. Subsequently, a particle swarm-optimized broad learning neural network is employed to rectify the SRCKF results from a compensatory standpoint, thereby enhancing the system’s filtering precision. Ultimately, the adaptive search window generated by Camshift(Continuously adaptive Mean shift) can be employed as an input for the credibility SRCKF, thereby facilitating the acquisition of more reliable real-time observation data. Finally, in the experimental part, the proposed algorithm is validated by the MNIST(Modified National Institute of Standards and Technology) dataset , CIFAR-10(Canadian Institute For Advanced Research) dataset, GTOT(Grayscale-Thermal Object Tracking) dataset and OTB2015(Object Tracking Benckmark 2015) dataset respectively for its generalization ability and robustness in static classification tasks and the effectiveness of dynamic target detection and tracking in real lighting scenes, Occlusion and motion blur scenes, while the performance is more advantageous than some mainstream algorithms.}, archive = {J_EAAI}, author = {Quanbo Ge and Zijian Xue and Hao Wang and Yuanliang Wang}, doi = {10.1016/j.engappai.2025.112165}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112165}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent multispectral image recognition by using credibility-based square root cubature kalman filters and broad particle swarm learning}, volume = {161}, year = {2025}, }
• (2025). Accidents analysis of mining industry through semantic text representation and dimensionality reduction: An integrated clustering framework. EAAI, 161, 112164. (www)

  @article{ , abstract = {Despite significant improvements in worker health and safety in recent decades, the mining industry still experiences fatal and non-fatal accidents. This underscores the critical need for a more nuanced understanding of accident causation patterns through accident data analysis. While conventional analytical approaches have yielded valuable insights, the extensive information embedded within text-based accident narratives remains underutilized. To address this gap, this study presents an artificial intelligence (AI)-based framework that integrates transformer-based natural language processing, nonlinear dimensionality reduction, and unsupervised machine learning to analyze and cluster accident narratives from the U.S. mining industry. Specifically, this study uses Sentence-BERT (SBERT), a sentence embedding model based on Bidirectional Encoder Representations from Transformers (BERT), to extract the high-dimensional semantical representation of accident narratives. These embeddings are then mapped to a low-dimensional space using the Uniform Manifold Approximation and Projection (UMAP) technique, followed by clustering with the k-means algorithm and subsequent hazard-focused cluster analysis. The primary contribution to AI lies in demonstrating the effectiveness of combining modern sentence embedding techniques with dimensionality reduction and clustering for the semantic analysis of safety-related narratives. From an engineering standpoint, this framework enables the identification of latent accident patterns that can inform hazard detection and guide safety interventions in the mining industry. The resulting clusters reveal diverse accident patterns across mining operations. In clusters associated with underground mining, a high proportion of incidents (ranging from 84 to 98 %) involved no equipment, with distinct injury patterns: torso injuries (67 %) from over-exertion, lower extremity injuries (58 %) from slips/falls, and upper extremity injuries from over-exertion (95 %) and material handling (85 %). Equipment-related clusters revealed strong associations with drilling tools (92 %), loaders (98 %), and bolting equipment (96 %). Clusters associated with strip/quarry/open pit operations exhibited a high frequency of vehicle-related accidents (98 % transportation, 99 % loaders), often resulting in multiple body part injuries. Milling operation clusters show 52–97 % no-equipment accidents, with injury patterns similar to those in underground mining. Additionally, noise-induced hearing loss (96–97 %) was observed in clusters spanning all mining operation types. These findings offer actionable insights for safety professionals and support data-driven, targeted interventions in the mining industry.}, archive = {J_EAAI}, author = {Abid Ali Khan Danish and Snehamoy Chatterjee and Kumar Vaibhav Raj and Rennie Kaunda and Hugh Miller and Aref Majdara}, doi = {10.1016/j.engappai.2025.112164}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112164}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Accidents analysis of mining industry through semantic text representation and dimensionality reduction: An integrated clustering framework}, volume = {161}, year = {2025}, }
• (2025). A lightweight and improved you only look once model using GhostWise convolution and attention mechanism for accurate plant disease detection. EAAI, 161, 112163. (www)

  @article{ , abstract = {Artificial intelligence enables early plant disease detection to support food security, yet achieving high accuracy with low computational cost on multi-class datasets remains challenging. To bridge this gap and solve the problem, we propose an improved version of the “You Only Look Once version 8 nano” (YOLOv8n) model that integrates a convolutional block attention module and custom coarse-to-fine modules into the baseline YOLOv8n architecture. To improve both efficiency and detection performance, the proposed model incorporates a combination of Ghost module and Depthwise convolution for lightweight feature extraction. Additionally, the attention modules emphasize important spatial and channel-wise features, helping the model focus better on critical information. The performance of the proposed model is compared with YOLO models including YOLOv5n, YOLOv5s, YOLOv7, YOLOv8n, and YOLOv8s. The experimental results show that the improved model outperforms the benchmark YOLO models. In addition, the improved model achieves mean Average Precision at Intersection over Union threshold of 0.5 ( mAP@0.5 ) 0.742, representing an improvement of 3.37 % over the baseline YOLOv8n. For the metric mean Average Precision at thresholds from 0.5 to 0.95 ( mAP@0.5:0.95 ), the improved model shows an improvement of 5.13 % over YOLOv8n. The precision measure reveals an improvement of 8.99 % and F1 score of 0.6856 outperforms YOLOv8n by 5.85 %. In addition, generalization experiments are conducted using commonly preferred plant disease datasets. These results validate the robustness, accuracy and stability of the improved model in effectively detecting plant leaf diseases.}, archive = {J_EAAI}, author = {Bunyamin Dikici and Mehmet Fatih Bekciogullari and Hakan Acikgoz and Serkan Ozbay}, doi = {10.1016/j.engappai.2025.112163}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112163}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and improved you only look once model using GhostWise convolution and attention mechanism for accurate plant disease detection}, volume = {161}, year = {2025}, }
• (2025). Hybrid optimization approach for syngas-fueled gas turbines: Integrating inverse problem solving and machine learning techniques. EAAI, 161, 112162. (www)

  @article{ , abstract = {This study investigates the efficiency of gas turbines powered by syngas derived from gasification, combining a detailed thermodynamic model with a hybrid optimization approach that integrates inverse problem solving via the Levenberg-Marquardt (LM) algorithm and supervised machine learning using Gradient Boosted Trees (GBT). The LM method is used to estimate turbine and compressor performance parameters based on efficiency maps extracted through a custom image processing routine. These maps are applied to calibrate isentropic efficiencies and define the system's operational boundaries. The resulting performance curves enhance the accuracy of the thermodynamic cycle model. To identify optimal operating conditions that maximize thermal efficiency and minimize entropy generation, the inverse problem formulation is integrated with the GBT model, enabling data-driven optimization based on simulated system behavior. All modeling and simulations are conducted in Wolfram Mathematica version 14.0, while verification is performed with the commercial software Gasturb version 14.0, using syngas composition data available in the literature. As a case study to demonstrate the method's applicability, syngas obtained from sewage sludge gasification is analyzed. The findings after optimization indicated an average thermal efficiency of 40 % when using syngas. The analysis revealed that the fuel mass flow rate contributes approximately 45 % to the efficiency gains and more than 70 % to the exergy reduction, while the excess air ratio contributes around 50 % to the cycle's efficiency. The study demonstrates the value of integrating LM-based modeling and GBT optimization, indicating significant potential for enhancing the performance of syngas-fueled turbines. Moreover, it highlights syngas from sewage sludge as a promising sustainable energy source.}, archive = {J_EAAI}, author = {Hiago David Zogbi Silva Oliveira and Vítor Caldeira de Andrada Bastos and York Castillo Santiago and Isabela Florindo Pinheiro}, doi = {10.1016/j.engappai.2025.112162}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112162}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid optimization approach for syngas-fueled gas turbines: Integrating inverse problem solving and machine learning techniques}, volume = {161}, year = {2025}, }
• (2025). Frequency-wavelet adaptive basis network for long-term time series forecasting. EAAI, 161, 112161. (www)

  @article{ , abstract = {Transformer-based models have demonstrated remarkable efficacy in long-term time series forecasting. Nevertheless, their fixed-scale feature extraction and static spectral processing limit adaptability to inherent multi-scale temporal dependencies. Additionally, conventional self-attention mechanisms struggle to effectively model hierarchical temporal structures. To address these limitations, we propose the Frequency-Wavelet Adaptive Basis Network (FWBNet), a novel framework that integrates three synergistic components including a Multi-scale Adaptive Basis Module (MS-ABM) for multi-resolution feature extraction, a Dual-Stream Wavelet Cross-Attention (DWCA) system with learnable wavelet transformations, and a computationally efficient Frequency-domain Multi-Layer Perceptron bottleneck (FreMLP bottleneck) for selective spectral processing. The MS-ABM innovatively integrates multi-scale features of time series through an adaptive basis function selection mechanism, dynamically fusing characteristics at different scales to form hierarchical representations. These multi-dimensional representations are then processed by the DWCA system, which establishes an efficient modeling framework for cross-scale temporal dependencies by combining bi-directional attention with learnable wavelet transforms. Building upon this foundation, the FreMLP bottleneck layer strategically performs selective feature optimization in the frequency domain, extracting key frequency information through spectral sparsification techniques to form an end-to-end multi-scale temporal feature learning pipeline. Experimental evaluations on six benchmark datasets demonstrate that FWBNet significantly outperforms state-of-the-art models, achieving mean squared error (MSE) reductions of 14% for multivariate and 21% for univariate predictions. This study presents a comprehensive framework for effective multi-scale temporal information integration in complex time series forecasting applications.}, archive = {J_EAAI}, author = {Qiang Lai and Yang You}, doi = {10.1016/j.engappai.2025.112161}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112161}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency-wavelet adaptive basis network for long-term time series forecasting}, volume = {161}, year = {2025}, }
• (2025). Balanced sample repository for knowledge distillation in data-free image classification scenario. EAAI, 161, 112160. (www)

  @article{ , abstract = {To address strict data security and privacy requirements in the field of knowledge distillation for image classification, generative network-based data-free knowledge distillation produces fake images for training student models to perform image classification tasks. However, we discover that the datasets consisting of those fake images often suffer from imbalances, such as unclear class margins, uneven intra-class distribution, and skewed label distribution, impeding classification accuracy. To this end, this paper proposes the balanced sample repository method for data-free knowledge distillation in the context of image classification, comprising three components: dynamic pseudo-supervised loss weight (DLW), biased label generator (BLG), and intra-class homogeneity operator (IHO). Specifically, DLW adjusts the pseudo-supervised loss weight to maintain distinct class margins, BLG uses the roulette algorithm to balance label distribution, and IHO employs local outlier factors to remove outliers and homogenize intra-class distributions. The fake images generated by the generative network are initially stored in the sample repository, and those three components achieve a balanced dataset by influencing this sample repository. The experiments conducted across five recognized image classification datasets demonstrate that the proposed method achieves competitive results and produces more balanced samples.}, archive = {J_EAAI}, author = {Yafeng Sun and Xingwang Wang and Junhong Huang and Shilin Chen}, doi = {10.1016/j.engappai.2025.112160}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112160}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Balanced sample repository for knowledge distillation in data-free image classification scenario}, volume = {161}, year = {2025}, }
• (2025). A lightweight detection model for green navel oranges in natural environments. EAAI, 161, 112157. (www)

  @article{ , abstract = {In natural environments, the detection of green navel oranges in the unstructured plantings of Gannan navel oranges presents challenges, including low model accuracy and excessive parameter size. This paper integrates existing technologies to propose a lightweight green tangerine detection model based on You Only Look Once version 5s (YOLOv5s). The aim is to improve the accuracy of green tangerine identification and detection in natural environments through model improvements. Firstly, the original YOLOv5 network model is optimized by replacing the original CSPDarknet53 (Cross Stage Partial Darknet53) backbone network with MobileNetV3 in order to reduce model parameters. Concurrently, the Squeeze-and-Excitation (SE) module in the network is substituted with the Convolutional Block Attention Module (CBAM), resulting in a lightweight network. Secondly, CBAM is integrated into the C3 network (cross-stage partial bottleneck with three convolutional layers) in the neck region to enhance the model's ability to extract green navel orange features after lightweighting. Finally, improve the loss function of the model to reduce the positioning error of the model for low-quality targets. The experimental results demonstrate that the enhanced model attains a 3.6 % enhancement in precision, 2.3 % in recall, and 2.1 % in mean Average Precision (mAP) in comparison with the original YOLOv5 model. The model's size is 3.9 megabytes (MB), representing a reduction of 72.9 % from the original 14.4 MB. Research shows that our method provides an efficient and accurate solution for identifying green navel orange targets in complex environments, with potential application value in agricultural automation for orchard management and monitoring.}, archive = {J_EAAI}, author = {Dongyuan Zhang and Qiusheng Li and Xindi Yuan}, doi = {10.1016/j.engappai.2025.112157}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112157}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight detection model for green navel oranges in natural environments}, volume = {161}, year = {2025}, }
• (2025). Efficient real-time instance segmentation of garment for intelligent robot tie-dye based on you only look once version 11 network. EAAI, 161, 112156. (www)

  @article{ , abstract = {Tie-dye, as an intangible cultural heritage with a long history, has gained global popularity due to its unique artistic value and versatile craftsmanship. However, its reliance on manual operation leads to low production efficiency and inconsistent quality. To address this, we propose an improved You Only Look Once version-11 (YOLOv11) instance segmentation model, Garment-YOLO, to efficiently segment the garment regions of robotic tie-dye by artificial intelligence technology. First, the C3k2_DynamicMixFormer (C3k2_DMF) module introduces dynamic kernel weight selection mechanism and multi-scale fusion to effectively balance local details and global information. Meanwhile, the Dual-Cross Recalibration Feature Pyramid Network (DCR-FPN) is proposed to enhance the detail preservation of edge region by selectively aggregating semantic and boundary information. Furthermore, the Superior-Head replaces depthwise convolution (DWConv) with part convolution (PConv), significantly reducing model complexity while maintaining performance. Experimental results demonstrate that Garment-YOLO achieves 167.4 frames per second (FPS) and maintains an optimal trade-off between inference speed and segmentation accuracy, reaching 91.9 % mean average precision (mAP)50–95 (Box), 91.5 % mAP50-95 (Mask). To further validate its practical performance, we conducted a 72-h production-line comparison, showing that compared to the baseline, we increase the product qualification rate by 24.6 %, reduce dye waste by 19.4 %, and increase the total production by 52.5 % compared to manual tie-dye. This study not only provides a feasible solution for the intelligent transformation of traditional tie-dye but also contributes to the preservation and development of this intangible cultural heritage. The code will be released on GitHub ( https://github.com/fudifu123/GARMENT-YOLO ). A video that intuitively introduces our research ( https://www.bilibili.com/video/BV13MgczAEkB ).}, archive = {J_EAAI}, author = {Difei Feng and Qihong Zhou and Lei Xiao and Kunfeng Ge and Hangzhou Ma and Li Zhou}, doi = {10.1016/j.engappai.2025.112156}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112156}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient real-time instance segmentation of garment for intelligent robot tie-dye based on you only look once version 11 network}, volume = {161}, year = {2025}, }
• (2025). A deep patch network with spatiotemporal meta-parameter learning for soft sensor modeling of industrial processes. EAAI, 161, 112155. (www)

  @article{ , abstract = {Soft sensing techniques have been widely employed in industrial processes to predict hardly measurable quality variables using easily measurable process variables. Effective modeling strategies for representing spatiotemporal features are highly desired in soft sensing applications for dynamic industrial processes. Deep networks, such as graph convolutional network and gated recurrent unit, are frequently employed to capture the spatial or temporal characteristics of sequential data. Nevertheless, effectively modeling and utilizing spatiotemporal heterogeneity remains a challenging problem. Thereby, a p atch network with spatiotemporal m eta- p arameter l earning (PMPL) is proposed in this paper, which presents an uniform framework of modeling spatiotemporal heterogeneous feature representations for soft sensing. In PMPL, a patching strategy is utilized to segment times series into sub-series-level patches, allowing the retention of local semantic information while capturing long-term dependencies. Next, a novel spatiotemporal meta-parameter learning approach is proposed to generate specific parameters for a spatial encoder, temporal encoder and spatiotemporal decoder through establishing corresponding embedding layers. In this way, the spatial, temporal and spatiotemporal dependencies within the process data can be adaptively captured and comprehensively refined from a uniform perspective. Extensive experiments are conducted based on two real production processes, which illustrate the feasibility and efficacy of the proposed model.}, archive = {J_EAAI}, author = {Xudong Shi and Kangping Du and Weili Xiong and Humberto Morales and Adriana Amicarelli}, doi = {10.1016/j.engappai.2025.112155}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112155}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep patch network with spatiotemporal meta-parameter learning for soft sensor modeling of industrial processes}, volume = {161}, year = {2025}, }
• (2025). Deep bioinspired evolutionary stacking algorithm for unpaired multimodal cell classification calibration. EAAI, 161, 112153. (www)

  @article{ , abstract = {Single-modality deep learning approaches for cell image classification exhibit inherent limitations in informational diversity when processing cross-institutional datasets acquired under varied imaging protocols. In contrast, multimodal cell imaging has emerged as a promising alternative for addressing data heterogeneity through comprehensive information integration. This study introduces a novel multimodal alternate training decision-making architecture based on a stacking algorithm for unpaired multimodal cell classification calibration. The method leverages deep bioinspired evolutionary networks combined with kernel-based support vector machines. Specifically, deep base classifiers incorporating multimodal concepts are derived from heterogeneous cell datasets. Each base classifier employs a bioinspired strategy to perform alternate training between two evolutionary densely connected attention networks. To mitigate class imbalance, where diseased cells are significantly outnumbered by normal cells, we incorporate a Shannon entropy loss term. Finally, multiple kernel-based support vector machines serve as meta classifiers, transforming high-level multimodal concepts into a separable feature space for robust decision-making. Experimental results demonstrate the superiority of our approach over existing algorithms for unpaired multimodal cell image classification. Our findings emphasize the importance of alternate training intra-modality classifiers and inter-modality fusion calibration for accurate and reliable medical image classification. All source code for this work will be publicly available on GitHub.}, archive = {J_EAAI}, author = {Lili Zhao and Di Xu and Xueping Tan and Jinzhao Yang and Weiping Ding and Hengde Zhu and Lichi Zhang and Qian Wang}, doi = {10.1016/j.engappai.2025.112153}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112153}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep bioinspired evolutionary stacking algorithm for unpaired multimodal cell classification calibration}, volume = {161}, year = {2025}, }
• (2025). Temporal-frequency joint hierarchical transformer with dynamic windows for speech emotion recognition. EAAI, 161, 112152. (www)

  @article{ , abstract = {Speech Emotion Recognition (SER) aims to identify the emotional state of a speaker from speech signals, serving as a critical prerequisite for achieving natural human–computer interaction. In speech signals, emotional information is inherently distributed across diverse frequency bands and temporal scales, with emotional cues in distinct regions exhibiting varying levels of heterogeneity or interdependence. Existing Transformer-based methods face limitations in precisely localizing salient temporal-frequency regions and modeling their inter-regional relationships. To address these challenges, a temporal-frequency joint hierarchical Transformer with dynamic window mechanisms, abbreviated as TF-DWFormer, is proposed to capture critical emotional cues and their contextual dependencies across temporal-frequency dimensions. It operates through several principal phases: Firstly, a feature reconstruction module is designed to extract temporal, frequency, and temporal-frequency representations of emotional speech. Secondly, a high-low frequency-based emotion-aware partitioning strategy is designed to achieve the division of emotional regions. Thirdly, a local window within a hierarchical Transformer analyzes static intra-region correlations to capture fine-grained emotional patterns, while a dynamic window adaptively models temporal evolution across regions, learning dynamic inter-region relationships. Lastly, a dual-cross-attention mechanism is employed to synergize comprehensive emotion representation from different domains. Our evaluation experiments demonstrate that the proposed TF-DWFormer method achieves recognition accuracies of 73.68%, 91.67%, 92.59%, 74.42%, and 50.54% on the datasets IEMOCAP, CASIA, EMODB, eNTERFACE05, and MELD, respectively, outperforming existing SER methods. These results confirm the capability of TF-DWFormer to precisely localize salient regions, robustly model inter-region dependencies, and effectively fuse multi-domain information, providing a promising solution for advancing SER technology.}, archive = {J_EAAI}, author = {Yonghong Fan and Heming Huang and Huiyun Zhang and Ziqi Zhou}, doi = {10.1016/j.engappai.2025.112152}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112152}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Temporal-frequency joint hierarchical transformer with dynamic windows for speech emotion recognition}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional reconstruction and fracture segmentation based on X-ray and computed tomography paired dataset. EAAI, 161, 112151. (www)

  @article{ , abstract = {In some orthopedic surgeries, the use of three-dimensional (3D) computed tomography (CT) scanning technology is not feasible due to scene limitations, leaving doctors to rely on two-dimensional (2D) X-ray images for real-time diagnosis. However, X-ray images lack 3D information, making accurate diagnosis challenging. Developing an algorithm to convert 2D X-ray images into 3D CT images, while simultaneously combining high-quality 3D reconstruction with precise fracture segmentation, offers a promising solution to the problem. In this study, we propose a novel artificial intelligence (AI)-driven framework named 3D reconstruction and segment anything model (3DRecSAM). The reconstruction image enhancer (RIE) is designed to achieve high-precision 3D reconstruction and provide high-quality feature initialization for fracture segmentation. Meanwhile, the mamba segment anything model (MSAM), based on the segment anything model (SAM) architecture, is developed for accurate fracture segmentation. We introduce a Kolmogorov–Arnold network (KAN)-based attention fusion module (KAF), which facilitates the joint optimization of the RIE reconstruction network and the MSAM segmentation network. Furthermore, the selective scanning mamba with KAN (SKM) is incorporated to enhance feature extraction for both RIE and MSAM. Mamba efficiently captures long-range dependencies and sequential patterns, while KAN’s learnable activation functions facilitate adaptive feature fusion and non-linear representation. To train and evaluate 3DRecSAM, we introduce the real X-ray and CT paired dataset (XCPData), which is publicly available on GitHub: https://github.com/YuanGao1201/XCPData .}, archive = {J_EAAI}, author = {Yuan Gao and Yuan Zhou and Da Chen and Jiachen Li and Mingle Zhou and Gang Li and Yunbo Gu and Jean-Louis Coatrieux and Yang Chen}, doi = {10.1016/j.engappai.2025.112151}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112151}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional reconstruction and fracture segmentation based on X-ray and computed tomography paired dataset}, volume = {161}, year = {2025}, }
• (2025). Unsupervised domain adaptation for lithology classification using dynamic entropy-based prototype learning. EAAI, 161, 112150. (www)

  @article{ , abstract = {Lithology classification plays a crucial role in geological exploration and resource evaluation. However, the significant distribution discrepancies between source and target domains, coupled with the unavailability of source domain data, pose substantial challenges to traditional domain adaptation methods. To address these challenges, we propose an innovative framework, Unsupervised Domain Adaptive Dynamic Entropy Prototype Learning (UDADEPL), which leverages a source-free unsupervised domain adaptation strategy for lithology classification. The UDADEPL framework consists of a frozen source pre-trained model and a trainable target model, incorporating a dynamic entropy-based prototype learning matrix for reliable sample selection and centroid-based pseudo-label learning for iterative optimization. Additionally, an information maximization loss and source domain regularization loss are integrated into a curriculum learning strategy to balance feature extraction and domain adaptation. This approach enables the model to effectively handle complex lithological boundaries and class imbalances in the target domain. Extensive experiments on datasets from the Tarim Oilfield and Hugoton–Panoma field demonstrate the superiority of UDADEPL over traditional machine learning and advanced deep learning models. UDADEPL achieves superior classification accuracy, outperforming the best baseline models, especially in cross-domain adaptation and complex lithology identification.}, archive = {J_EAAI}, author = {Hengxiao Li and Yahui Liu and Lu Liu}, doi = {10.1016/j.engappai.2025.112150}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112150}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised domain adaptation for lithology classification using dynamic entropy-based prototype learning}, volume = {161}, year = {2025}, }
• (2025). Transforming physics-informed machine learning to convex optimization. EAAI, 161, 112149. (www)

  @article{ , abstract = {Physics-Informed Machine Learning (PIML) offers a powerful paradigm of integrating data with physical laws to address important problems in engineering, such as parameter estimation, inferring hidden physics, equation discovery, and state prediction. However, PIML, such as Physics-Informed Neural Networks (PINNs), still lack the necessary accuracy, stability, and interpretability when applying in practical engineering due to many serious optimization challenges including the spectral bias, non-convex optimization, multi-objective optimization, and non-smooth optimization. In this study, we propose the Convex-PIML based on convex optimization and basis functions widely used in well-established numerical solvers to overcome all these limitations. The linear combination of B-splines is utilized to approximate the data, promoting the convexity of the loss function. By dividing variables into blocks and replacing the non-convex loss terms with convex approximations, the problem is further converted into a sequence of successively refined approximated convex optimization problems. This conversion known as Block Successive Convex Approximation (BSCA) allows the use of well-established convex optimization algorithms, obtaining solutions effectively and efficiently. Furthermore, an adaptive knot optimization method is introduced to mitigate the spectral bias issue of PIML, further improving the performance. The proposed fully adaptive framework by combining the adaptive knot optimization and BSCA is tested in scenarios with distinct types of physical prior. The results indicate that optimization problems are effectively solved in these scenarios, highlighting the potential of the framework for broad applications. Note that the Convex-PIML is also flexible since many other basis functions can also be incorporated to handle different systems.}, archive = {J_EAAI}, author = {Letian Yi and Siyuan Yang and Ying Cui and Zhilu Lai}, doi = {10.1016/j.engappai.2025.112149}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112149}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transforming physics-informed machine learning to convex optimization}, volume = {161}, year = {2025}, }
• (2025). Finite-time fuzzy tracking control for networked suspension systems under gilbert-elliott fading channels. EAAI, 161, 112148. (www)

  @article{ , abstract = {In this paper, we address the finite-time ℓ 2 − ℓ ∞ fuzzy tracking control problem for networked suspension systems operating under Gilbert-Elliott fading channels. Firstly, the suspension system is established based on interval type-2 fuzzy rules to effectively describe the nonlinearity and uncertainty inherent in the model. Secondly, the Gilbert-Elliott model is employed to characterize the random switching behavior of channel states (good channel and fading channel) under real-world conditions. The timely identification of the current channel state is achieved through a mode detection method. To address channel fading and communication constraints, a novel mode-dependent controller is proposed. Simultaneously, we aim to minimize unnecessary and redundant communication. The hybrid scheduling strategy of the FlexRay protocol is utilized to transmit the measurement output data to the observer, and a saturation function is introduced in the observer to minimize the impact of outliers. Subsequently, the design conditions for stochastic finite-time boundedness and ℓ 2 − ℓ ∞ performance of the augmented system is proposed. Finally, simulation results demonstrate that the proposed algorithm is practically viable in dealing with complex environments, such as Gilbert-Elliott fading channels, outliers, and communication constraints.}, archive = {J_EAAI}, author = {Li-Juan Cai and Xiao-Heng Chang and Xin-Yue Zhao}, doi = {10.1016/j.engappai.2025.112148}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112148}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finite-time fuzzy tracking control for networked suspension systems under gilbert-elliott fading channels}, volume = {161}, year = {2025}, }
• (2025). RT-FedFlow: An efficient framework for real-time traffic signal optimization using federated multi-agent reinforcement learning. EAAI, 161, 112147. (www)

  @article{ , abstract = {The optimization of Traffic Signal Control (TSC) is a critical component of the Intelligent Transportation System (ITS) in smart cities, aiming to eliminate traffic congestion and improve urban mobility. However, traditional approaches to TSC often rely on centralized data processing, limiting scalability and adaptability to real-time traffic dynamics while raising concerns about data privacy, communication, and computation. To address these challenges, this paper proposes the RT-FedFlow, a novel federated multi-agent reinforcement-learning (FMARL) framework for real-time traffic signal optimization. The framework employs enhanced FMARL techniques to enable decentralized collaboration among multiple traffic signal agents, optimizing traffic flow while seamlessly balancing local adaptability and global coordination. Federated model aggregation enables agents to collaboratively learn a global policy while preserving local data privacy, with each reinforcement learning (RL) agent independently optimizing its policy. Next, an advanced reward mechanism is designed to minimize congestion and reduce waiting times. Moreover, a hierarchical communication and coordination strategy is adopted to facilitate efficient local interaction among agents and global model aggregation to improve overall system performance. In addition, the framework enables real-time optimization and prioritization of emergency vehicles through a dedicated preemption module. Compared to existing TSC models, RT-FedFlow significantly improves traffic signal coordination and response times by dynamically adapting to complex, real-time traffic conditions. The proposed RT-FedFlow framework is implemented and evaluated using the SUMO simulator. Extensive simulation results demonstrate that RT-FedFlow outperforms all benchmark models, achieving a 15.49 % reduction in average inference time, a 3.07 % reduction in average queue length, a 13.04 % reduction in average intersection delay, a 0.20 % improvement in fuel consumption, and a 5.35 % emergency vehicle clearance time.}, archive = {J_EAAI}, author = {Ayaz Akbar and Syed Sajid Ullah and Abdul Malik and Saeed Mian Qaisar}, doi = {10.1016/j.engappai.2025.112147}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112147}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {RT-FedFlow: An efficient framework for real-time traffic signal optimization using federated multi-agent reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). A hybrid data-driven model for state of health estimation of lithium-ion battery with capacity recovery. EAAI, 161, 112146. (www)

  @article{ , abstract = {The failure of lithium-ion batteries has attracted the attention of researchers. Monitoring the degradation process through sensor-based state-of-health (SOH) assessment enables early detection of anomalies and facilitates timely interventions to prevent catastrophic failures. During battery degradation, the phenomenon of capacity recovery makes it complicated to accurately evaluate the SOH. However, existing works usually neglect this phenomenon. In this paper, we propose a hybrid model focusing on capacity recovery. The model consists of four modules: a feature extraction module, a data decomposition module, a global trend assessment module, and a local fluctuation assessment module. Specifically, in the feature extraction module, a convolutional neural network extracts features from the input data that effectively characterize the capacity degradation process. The data decomposition module applies empirical mode decomposition (EMD) to separate local fluctuations caused by capacity recovery from the global degradation trend. These components are then input into their respective assessment modules. Additionally, the local fluctuation assessment module incorporates an attention mechanism that adaptively identifies the similarity between extracted features and health states, thereby enhancing the evaluation of local fluctuations. The proposed model is validated using the NASA (National Aeronautics and Space Administration) battery dataset, demonstrating its effectiveness in addressing the challenges of SOH prediction under capacity recovery conditions.}, archive = {J_EAAI}, author = {Yu Lin and Luo Zhou and Jianhai Yan and Shuguang He}, doi = {10.1016/j.engappai.2025.112146}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112146}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid data-driven model for state of health estimation of lithium-ion battery with capacity recovery}, volume = {161}, year = {2025}, }
• (2025). A novel localization algorithm integrated with sensor fusion for position estimation of mobile robot in the road following and roundabout environment. EAAI, 161, 112145. (www)

  @article{ , abstract = {A novel Artificial Intelligence algorithm based autonomous mobile robot localization in road environments is presented in this paper. The sensor fusion data, employing camera and odometry, are used to build the local maps for road environment. The localization of the robot is performed using a novel Artificial Intelligence algorithm, called Priori-Posteriori-Laser-Simulator, to find the position and heading-angle of robot along movement in two phases, namely priori and posteriori stages as 1st phase and Kalman filter in 2nd phase. In the priori stage, the path of the robot is planned in the built-map using Laser Simulator through generating a series of points as lines to drive the robot in the middle of the road environments. In the posteriori-stage, the position and robot heading angle are measured using sensor fusion system and the laser-simulator updates the position accordingly through a comparison between the planned and actual positions. The Kalman filter is used in 2nd phase to remove noise and enhance to enhance the localization accuracy. The proposed algorithm has been tested in indoor and outdoor road environments to show its effectiveness and performance for localizing robot in such environments. It is also compared with fuzzy-logic, adaptive H-infinity filter, fuzzy logic with Kalman filter and Oriented-Rotated Binary Simultaneous Localization and Mapping 2 algorithms on localizing robot at road environment. Results show the capability of the proposed algorithm to enable the robot to move effectively in road environments, recognize the features of road terrains and localize the robot during autonomous navigation in road environments.}, archive = {J_EAAI}, author = {Mohammed A.H. Ali and Nukman Yusoff and Bushroa Abd Razak and Sherzod Turaev and Rawad Abdulghafor and Aisha Muhammad}, doi = {10.1016/j.engappai.2025.112145}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112145}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel localization algorithm integrated with sensor fusion for position estimation of mobile robot in the road following and roundabout environment}, volume = {161}, year = {2025}, }
• (2025). Electroencephalography-based emotion recognition using a dual-stream multi-scale spatiotemporal convolutional capsule network. EAAI, 161, 112144. (www)

  @article{ , abstract = {In recent years, emotion recognition using electroencephalogram (EEG) signals has gained significant attention. However, shallow convolutional neural networks used in EEG emotion recognition are ineffective at capturing spatial relationships between features, adversely affecting model performance. To solve this problem, this study proposes a dual-stream multi-scale spatiotemporal convolutional capsule network aimed at improving EEG-based emotion recognition. The novelty of our approach is the dual-stream multi-scale spatiotemporal design, which enables parallel extraction and fusion of temporal and spatial EEG features at multiple scales, offering a richer and more discriminative representation for emotion recognition. Specifically, a dual-stream feature construction module is developed to extract multi-scale temporal and spatial features from raw EEG signals. A hybrid spatiotemporal attention mechanism enhances feature fusion, while a capsule-based classifier improves recognition accuracy by modeling relationships between local and global features. Experimental results using subject-dependent 10-fold cross-validation show average accuracies of 97.72%, 97.56%, and 97.82% for valence, arousal, and dominance on the Dataset for Emotion Analysis using Physiological signals(DEAP), with average F1 scores of 97.83%, 97.51%, and 97.68%. For the Dataset for Emotion Analysis using Physiological signals(DREAMER 1 ), the average accuracies are 96.48%, 96.32%, and 96.35%, with corresponding F1 scores of 96.23%, 95.74%, and 95.66%. The proposed method outperforms existing state-of-the-art approaches on both datasets, reducing the number of parameters by 58.07% and decreasing inference time by approximately 50.86% and 33.39%, compared to Residual Network 18. Additionally, in the subject-independent leave-one-subject-out cross-validation, the proposed method demonstrated results that significantly outperformed the baseline model across both datasets. Experiments show that the proposed method enhances spatiotemporal feature extraction and improves emotion recognition accuracy. This method reduces computational resource consumption and enhances recognition accuracy, thereby facilitating efficient algorithm development and deployment for real-time emotion monitoring and related applications.}, archive = {J_EAAI}, author = {Han Cai and Pengfei Lu and Xiaofang Wang and Yuxing Wang}, doi = {10.1016/j.engappai.2025.112144}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112144}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Electroencephalography-based emotion recognition using a dual-stream multi-scale spatiotemporal convolutional capsule network}, volume = {161}, year = {2025}, }
• (2025). Inverse structural design with generative and probabilistic autoencoders and diffusion models. EAAI, 161, 112143. (www)

  @article{ , abstract = {Traditional structural design is a forward trial-and-error process. Designers need to iterate through different design solutions and conduct structural analysis until the design meets the codes and standards. This study proposes and investigates a generative machine learning (ML) framework for inverse design of continuous beam systems. Three generative ML models, including conditional variational autoencoder (CVAE), conditional autoencoder with maximum likelihood estimation (CAE-MLE), and denoising diffusion models (DDMs) are trained and fine-tuned on the CBeamXP (Continuous Beam Cross-section Predictors) dataset with 1,000,000 beam sections to generate the cross sectional properties. Research results show that CAE-MLE achieves the highest generation accuracy and robustness, while CVAE offers more variability through latent space sampling. DDMs provide controllable generation variability via a stochasticity parameter in the inverse diffusion process. The proposed framework enables efficient generation of multiple design solutions and can potentially accelerate the conceptual design workflows in structural engineering. This work also demonstrated the feasibility toward artificial intelligence (AI)-assisted structural design using generative approaches and tabular datasets.}, archive = {J_EAAI}, author = {Bozhou Zhuang and Adrien Gallet and Danny Smyl}, doi = {10.1016/j.engappai.2025.112143}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112143}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inverse structural design with generative and probabilistic autoencoders and diffusion models}, volume = {161}, year = {2025}, }
• (2025). Multi-objective generative design framework and realization for quasi-serial manipulator: Considering kinematic and dynamic performance. EAAI, 161, 112142. (www)

  @article{ , abstract = {This paper proposes a framework for optimizing the linkage mechanism of a quasi-serial manipulator for target tasks. The process is illustrated through a case study of a two-degree-of-freedom (2-DOF) linkage mechanism, which significantly influences the workspace of the quasi-serial manipulator. First, a diverse set of quasi-serial mechanisms is generated with workspaces that satisfy the target task and is converted into three-dimensional computer-aided design (3D CAD) models. Then, kinematic and dynamic analyses are conducted to compute workspace and payload torque labels for surrogate model training. Adaptive sampling is employed to identify an appropriate amount of training data for accurate prediction, while ensemble models are utilized to enhance robustness. After model training, a multi-objective optimization problem is formulated under the mechanical and dynamic constraints of the manipulator. The design objective is to recommend quasi-serial mechanisms with optimized kinematic (workspace) and dynamic (payload torque) performance that fulfill the target task. To explore the underlying physics of the Pareto solutions obtained via the Non-Dominated Sorting Genetic Algorithm (NSGA-II), various data mining techniques are applied to extract design rules that offer practical guidance. Finally, a detailed manipulator is realized using 3D-printed parts with topology optimization, and its performance is verified through a payload test. These results demonstrate the potential of the proposed framework for broader mechanism applications and its capability to support practical design decisions through design rule extraction.}, archive = {J_EAAI}, author = {Sumin Lee and Sunwoong Yang and Namwoo Kang}, doi = {10.1016/j.engappai.2025.112142}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112142}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective generative design framework and realization for quasi-serial manipulator: Considering kinematic and dynamic performance}, volume = {161}, year = {2025}, }
• (2025). Blockchain and deep learning-enabled IoT device-to-device authentication approach for smart cities using 5th generation technology. EAAI, 161, 112141. (www)

  @article{ , abstract = {As smart cities evolve, the increasing number of Internet of Things (IoT) devices requires secure authentication mechanisms for device-to-device (D2D) communication, especially within 5G networks. To address security challenges in D2D communication, Blockchain (BC) technology is utilized. Existing methods face issues like single-point failure attacks, high implementation costs, and data privacy concerns. This work aims to develop a secure and efficient authentication model leveraging BC technology and a Deep Q Network (DQN) for key generation to enhance IoT security in smart city applications. A novel authentication model, Deep Q Network-SecAuth (DQN-SecAuth), is proposed for secure D2D communication using BC. The model involves key entities such as the Registration Authority Center (RAC), blockchain, and IoT devices. The authentication process includes six stages: setup, registration, key generation, authentication, new device addition, and formal verification. In the key generation stage, the secret key is generated using a Deep Learning (DL) model named DQN. The proposed model also incorporates encryption, hash functions, Chebyshev polynomials, and XOR operations to ensure security. The DQN-SecAuth model achieved a minimum computational time of 9.525 s and memory usage of 42.897 megabytes (Mb), consensus delay of 0.420 s, energy consumption of 0.715 J, latency of 0.469 s, power consumption of 0.270 kW-hour (kWh), and throughput of 0.804 megabits per second (Mbps). The key novelty of this work is the use of a DQN for adaptive key generation, integrated into a BC-based authentication framework. This combination enhances security, reduces computational overhead, and outperforms traditional static key-generation and authentication methods in smart city IoT environments.}, archive = {J_EAAI}, author = {K. Prabhu Chandran and Bhuvaneswari P and Sivasankaran V and Vimala S}, doi = {10.1016/j.engappai.2025.112141}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112141}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blockchain and deep learning-enabled IoT device-to-device authentication approach for smart cities using 5th generation technology}, volume = {161}, year = {2025}, }
• (2025). Emergency medical facility site selection in drone-based relief operations using an enhanced T-spherical fuzzy frank combined compromise solution method. EAAI, 161, 112140. (www)

  @article{ , abstract = {Unmanned aerial vehicles (UAVs) play a vital role in disaster response operations, emphasizing the need for efficient site selection for emergency medical facilities (EMFs). This study presents a structured framework for evaluating candidate sites and identifying critical criteria for establishing EMFs in drone-based relief operations. It proposes a novel T-spherical fuzzy (T-SF) multi-criteria group decision-making framework for EMF site selection. The framework integrates the T-SF-Entropy method and a hybrid subjective–objective weighting scheme that combines stepwise weight assessment ratio analysis and the method based on the removal effects of criteria. This integration enables a more reliable aggregation of expert opinions. The framework applies T-SF Frank aggregation operators in the modified combined compromise solution method to rank potential sites under multi-disaster conditions. A real-world case study evaluates alternative sites using the critical criteria and demonstrates the practical utility of the proposed model in the field of emergency engineering logistics. Quantitative analysis shows that the “urban logistics hub” achieves the highest compromise index due to its advantages in proximity, infrastructure, and supply chain access. Sensitivity analysis confirms that variations in parameters do not affect the stability of rankings. This study uses artificial intelligence to support intelligent decision-making in disaster response. It provides an effective engineering application that optimizes UAV-based healthcare deployment, enhances resource allocation, and improves the speed and accuracy of emergency medical responses.}, archive = {J_EAAI}, author = {Rajdip Mahajan and Saptadeep Biswas and Vladimir Simic and Dragan Pamucar and Abhijit Baidya and Uttam Kumar Bera}, doi = {10.1016/j.engappai.2025.112140}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112140}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Emergency medical facility site selection in drone-based relief operations using an enhanced T-spherical fuzzy frank combined compromise solution method}, volume = {161}, year = {2025}, }
• (2025). Feature description attention: Channel-independent local–global fusion for multi-scale feature representation. EAAI, 161, 112139. (www)

  @article{ , abstract = {In recent deep convolutional networks, attention mechanisms rely heavily on global information to generate attention weights, but few methods can effectively recalibrate global features based on local object characteristics. To address this gap, we introduce a masked-averaging strategy that adaptively selects regions of interest from the feature map, allowing local features to encode and reflect global information. By combining these local descriptors with global averages and maximums into a multi-scale Bag-of-Visual-Words (BoVW), our method jointly captures salient point, local region, and global context, resulting in richer and more discriminative feature representations. Additionally, we propose Feature Batch Normalization (FBN) to facilitate cross-channel interactions, further enhancing the performance of attention modules. In terms of engineering applications, to overcome the limitations of traditional channel-dependent attention under model compression and pruning, we design a channel-independent feature recalibration mechanism based on one-dimensional depthwise convolution, termed Feature Description Attention (FDA). It leverages the properties of BoVW and FBN to provide a flexible and lightweight attention module with minimal computational overhead and robust performance under quantization and pruning. Our method demonstrates consistent gains across three vision tasks: classification accuracy improves by 0.72–1.68%, object detection performance increases by 1.8–3.0% mean average precision, and segmentation accuracy rises by 1.9%. These results across different benchmarks confirm the effectiveness and generalizability of our approach. FDA offers a scalable, hardware-friendly solution for modern vision applications, especially valuable in resource-constrained settings such as mobile deployment, autonomous systems, and medical image analysis.}, archive = {J_EAAI}, author = {Yuanyang Zhu and Guangjie Han and Hongbo Zhu and Fan Zhang}, doi = {10.1016/j.engappai.2025.112139}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112139}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature description attention: Channel-independent local–global fusion for multi-scale feature representation}, volume = {161}, year = {2025}, }
• (2025). Improving robustness of transformers for power quality disturbance classification via optimized relevance maps. EAAI, 161, 112138. (www)

  @article{ , abstract = {Power quality disturbances (PQDs) are a critical area of research in the power systems domain. The growing complexity of modern power systems increases the variability of PQDs that may occur, making it challenging to manage power quality using traditional methods, and motivates the inclusion of machine learning models. However, these models are vulnerable to adversarial attacks. The predominant defense approach is adversarial training, which increases robustness at the price of significant increase in computational complexity of the training process, and can often lead to reduced accuracy on non-adversarial examples. In this light, this paper introduces a distinct approach to increase the robustness of Transformer-based PQD classifiers by leveraging their intrinsic explainability derived from the self-attention mechanism. By incorporating the resulting explainable features in the form of relevance maps into the loss function during the fine-tuning stage, we guide the model to focus on salient disturbance features. This method enhances the robustness against adversarial attacks, without compromising accuracy, and avoids the additional computational overhead associated with adversarial training. Experimental results show that when no adversarial attack is applied, the fine-tuning stage increases the PQD classification accuracy from 97.62% to 99.26%, thus overcoming the accuracy-robustness trade-off. Furthermore, the Transformer with the proposed fine-tuning stage is more adversarial robust than a state-of-the-art deep Convolutional neural network (DeepCNN) PQD classifier. The proposed Transformer maintains accuracy and F1 scores slightly above 88%, whereas fo DeepCNN both scores are slightly below 66.60%, thus highlighting the effectiveness of our approach.}, archive = {J_EAAI}, author = {Itamar Kapuza and Elinor Ginzburg-Ganz and Ram Machlev and Yoash Levron}, doi = {10.1016/j.engappai.2025.112138}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112138}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving robustness of transformers for power quality disturbance classification via optimized relevance maps}, volume = {161}, year = {2025}, }
• (2025). Progressive epipolar geometry for robust light field super-resolution. EAAI, 161, 112137. (www)

  @article{ , abstract = {Light field technology captures both spatial and angular information, densely sampled high-resolution light field images contain abundant 3-dimensional geometric information, enabling widespread applications in various industrial fields. However, existing methods for joint spatial and angular super-resolution of light field images face significant challenges when reconstructing scenes with large disparities and significant occlusions. To address this issue, we propose a geometric information enhancement module that enables efficient extraction and preservation of geometric information, and then we propose a non-disparity-based, one-stage approach that allows for the direct reconstruction of densely sampled high-resolution light field images from sparsely sampled low-resolution counterparts. Specifically, we decompose the original 4-dimensional light field image into three distinct 2-dimensional representations: spatial, angular, and epipolar plane image. Among these representations, the epipolar plane image contains abundant geometric information, which inspires us to propose a more efficient feature extractor. In addition, we introduce a progressive feature fusion strategy that better preserves geometric information extracted from epipolar plane images. Finally, to avoid errors introduced by warping operations that complicate the process of enhancing geometric information, we introduce a spatial-angular integrated upsampling module. Extensive experimental results on public datasets demonstrate that our proposed method significantly outperforms state-of-the-art approaches both quantitatively and qualitatively. Specifically, on the Occlusions dataset, our method achieved significant improvement in performance while reducing inference time by approximately 80% compared to the current best method. This efficiency gain is particularly beneficial for practical applications of the artificial intelligence algorithm.}, archive = {J_EAAI}, author = {Hao Zhang and Hao Sheng and Rongshan Chen and Da Yang and Ruixuan Cong and Zhenglong Cui and Xuefei Huang and Guanqun Su}, doi = {10.1016/j.engappai.2025.112137}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112137}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Progressive epipolar geometry for robust light field super-resolution}, volume = {161}, year = {2025}, }
• (2025). Exploring knee joint moment prediction models for landing task with various model architecture: Leveraging inertial measurement unit sensors data and subject-specific attributes. EAAI, 161, 112136. (www)

  @article{ , abstract = {This study presents a comparative evaluation of deep learning models for predicting Three-Dimensional (3D) knee joint moments during landing tasks using Inertial Measurement Unit (IMU) data. We assess a range of architectures, from simple feedforward networks to complex Three-Dimensional Convolutional Neural Networks (3D-CNN), to balance predictive accuracy with computational cost. The 3D-CNN model achieved the highest performance, with relative Root Mean Squared Error (rRMSE) values of 6.08 % for the knee abduction moment, 5.3 % for the external rotation moment, and 4.9 % for the flexion moment. We further explored the effect of adding subject-specific features, which yielded modest performance gains. While these features provided some improvement, model performance primarily depended on sensor data, reinforcing the need to integrate both biomechanical and Artificial Intelligence-driven insights. From an engineering application perspective, our findings guide researchers and clinicians in selecting the most suitable model and sensor placement based on both accuracy and computational complexity. Specifically, we highlight thigh-mounted IMU as the most effective placement for capturing landing dynamics. These insights support the practical deployment of knee joint moment prediction systems in real-world settings, such as gyms or training centers, where compact sensors and limited computational resources are essential.}, archive = {J_EAAI}, author = {Tommy Sugiarto and Yi-Jia Lin and Ya-Wen Tu and Hsiao-Liang Tsai and Lin-Fen Hsieh and Chi-Tien Sun and Patrik Kutilek and Wei-Chun Hsu}, doi = {10.1016/j.engappai.2025.112136}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112136}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring knee joint moment prediction models for landing task with various model architecture: Leveraging inertial measurement unit sensors data and subject-specific attributes}, volume = {161}, year = {2025}, }
• (2025). Key region semantic information augmented transformer for image captioning. EAAI, 161, 112135. (www)

  @article{ , abstract = {Existing image captioning models often face difficulties in capturing inter-object relationships and generating description that comprehensively understands the entire image content, either relying on object detectors that overlook contextual information or depending on grid features that fail to adequately model spatial interactions. This paper proposes two solutions to these challenges. The first is the introduction of a module for mining semantic information from key regions. Based on the spatial proximity and high co-occurrence between objects, this module identifies the public region covered by these objects as a key region, mines their semantic information, and incorporates it into the modeling process, which compensates for the limitations of grid features. Second, we improve the standard Transformer decoder’s architecture by innovatively introducing an adaptive gating mechanism that dynamically adjusts the alignment between textual and visual features, enhancing the model’s overall comprehension of the image. To validate our approach, we applied these modules to the Transformer framework and proposed a novel method for image captioning, called Key region Semantic information Augmented Transformer (KSAT) for Image Captioning. Extensive experiments on benchmark datasets show that the proposed method outperforms many models. Specifically, our method achieves a score of 139.6% on the offline test, and 138.4% on the official online test server on the Consensus-based Image Description Evaluation (CIDEr) metric. In qualitative evaluation, our method also outperforms other methods at generating captions for complex scenes. Overall, these results confirm the validity of our method and advance the field of artificial intelligence.}, archive = {J_EAAI}, author = {Fuyun Deng and Wei Li and Zhixin Li}, doi = {10.1016/j.engappai.2025.112135}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112135}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Key region semantic information augmented transformer for image captioning}, volume = {161}, year = {2025}, }
• (2025). Enhanced neural-network-based iterative learning control considering iterative uncertainties for piezoelectric actuated micro-positioning platform. EAAI, 161, 112134. (www)

  @article{ , abstract = {This research aims to developing a new enhanced data-driven sliding-mode iterative learning control (E-DDSILC) strategy for piezoelectric actuated micro-positioning (PAMP) platforms. For the first time, the analysis demonstrating that errors converge to 0 in E-DDSILC is successfully extended from strictly repetitive systems to systems with non-strictly repetitive initial conditions. This generalization expands the practical application range of E-DDSILC. Simultaneously, iterative uncertainties are considered, which are the major factor affecting the performance of iterative learning control. To address these uncertainties, a diagonal recurrent neural network is employed to fit and compensate for them within a dynamic linearization model, thereby further enhancing the tracking accuracy and practicability of E-DDSILC. Finally, Several experiments are performed on a PAMP platform to compare the developed E-DDSILC method with both classical DDSILC and traditional E-DDSILC schemes. Comparative experimental results prove the superiority of the developed controller.}, archive = {J_EAAI}, author = {Miaolei Zhou and Yulong Sun and Xiuyu Zhang and Wei Gao and Chun-Yi Su}, doi = {10.1016/j.engappai.2025.112134}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112134}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced neural-network-based iterative learning control considering iterative uncertainties for piezoelectric actuated micro-positioning platform}, volume = {161}, year = {2025}, }
• (2025). A data-driven multi-stage stochastic optimization for sustainable humanitarian supply chain using machine learning algorithms. EAAI, 161, 112133. (www)

  @article{ , abstract = {The frequency and severity of natural disasters have intensified, resulting in significant human, financial, and emotional consequences. Earthquakes, in particular, have caused severe economic losses, deaths, and homelessness among millions. This study is designed to establish a comprehensive plan for managing pre- and post-disaster phases, including preparations, responses, and recovery efforts. It introduces a Multi-Stage Stochastic Programming (MSSP) model for sustainable humanitarian relief operations, optimizing location, allocation, and inventory management. The first and third stages concentrate on minimizing environmental impacts, while the second stage centers on enhancing social welfare. Simultaneously, economic cost reduction is consistently pursued in all three stages. The model's primary advantages include optimized inventory management to avoid shortages and flexible logistics strategies for timely and cost-effective delivery of relief items. Additionally, it ensures continuous aid, addressing both short-term and long-term needs to improve disaster management effectiveness and resilience. The multi-objective model is solved using Augmented Epsilon-Constraint (AEC). Furthermore, this paper employs Multi-Criteria Decision-Making (MCDM) methods to rank suppliers, leveraging Machine Learning (ML) algorithms to enhance ranking precision, thereby leading to a more responsive Supply Chain (SC). A real-world case study is then provided to illustrate the applicability and validity of the proposed model. Focusing on achieving balanced sustainability across all three stages, ensuring seamless logistics for all humanitarian supplies and affected individuals, and addressing uncertainties, the model determines the optimal quantities of all relief items to store. Through comprehensive analysis, the results provide key insights into the importance of MSSP in disaster management plans, enhancing understanding of the model's effectiveness.}, archive = {J_EAAI}, author = {Farnaz Ansari and Ali Bozorgi-Amiri and Hossein Shakibaei}, doi = {10.1016/j.engappai.2025.112133}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112133}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A data-driven multi-stage stochastic optimization for sustainable humanitarian supply chain using machine learning algorithms}, volume = {161}, year = {2025}, }
• (2025). A theme music generation model based on hybrid variational autoencoders and conditional generative adversarial networks. EAAI, 161, 112131. (www)

  @article{ , abstract = {The musical theme is the core melodic element that runs through a work. It is the main source of recognizable features and the basis of the overall structure of the score. However, existing music generation systems often face problems with missing thematic features and monotonous structure. These systems find it difficult to capture the complex thematic changes and long-term dependencies in music, resulting in a lack of dynamic changes, artistic expression in phrasing, and structural coherence in the generated results. In this paper, we propose a thematic music generation model based on a generative adversarial network, named Thematic Music Conditional Generative Adversarial Network (TM-CGAN), which solves these challenges through three key innovations. First, we propose a thematic fusion degree calculation module based on multi-dimensional feature metrics to overcome the theoretical limitations of existing methods, which lack explicit modeling of thematic features. Subsequently, we construct a two-dimensional image representation learning framework that preserves Musical Instrument Digital Interface (MIDI) symbolic sequences and employs convolutional neural networks (CNNs) to effectively model the multi-dimensional feature dependencies inherent in musical structures. Finally, we design a hybrid variational autoencoder-conditional generative adversarial network architecture that processes latent space modeling of thematic features through variational inference mechanisms while simultaneously optimizing generation quality through adversarial training. We conducted comprehensive experiments across three diverse MIDI music datasets to validate our approach. The experimental results demonstrate that TM-CGAN significantly outperforms state-of-the-art baseline models on multiple evaluation metrics, including generated music quality, theme representativeness, and structural integrity.}, archive = {J_EAAI}, author = {Fangzhu Jin and Peng Li and Xiaojun Wu}, doi = {10.1016/j.engappai.2025.112131}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112131}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A theme music generation model based on hybrid variational autoencoders and conditional generative adversarial networks}, volume = {161}, year = {2025}, }
• (2025). Malicious detection and trust calculation using residual recurrent neural network for trust with quality of service-aware multicast routing in mobile ad-hoc network system. EAAI, 161, 112130. (www)

  @article{ , abstract = {A Mobile Ad-hoc Network is a collection of mobile nodes without any proper infrastructure. In this work, a novel trust and Quality of Service-aware multicast routing technique for Mobile Ad-hoc Network is introduced. The key scope of this research paper is to evaluate the trust with Quality of Service -aware multicast routing process in Mobile Ad-hoc Network by detecting malicious nodes. For performing the optimal routing without any suspicious attacks initially the malicious node detection is performed by Residual Recurrent Neural Network. Further, if the Mobile Ad-hoc Network model is normal and free from malicious nodes, then the true value is calculated by utilizing the outcome of malicious node detection. If the node is free from malicious then, the trust value becomes high or else the trust value becomes low. Once, the trust calculation is completed the optimal routing is performed in the Mobile Ad-hoc Network with the support of the Enhanced Artificial Rabbits Optimization algorithm. Moreover, different constraints like hop count, throughput, Packet Delivery Ratio, delay, and energy consumption are derived. Further, different experiments are evaluated to prove the effectiveness of the implemented model against several baseline technique. Hence, the developed model accomplishes superior efficiency in detecting the malicious and performs the multicast routing in the Mobile Ad-hoc Network.}, archive = {J_EAAI}, author = {Sanjaya Kumar Sarangi and Rasmita Lenka and Janmejaya Mishra and Ritarani Sahu and Arabinda Nanda}, doi = {10.1016/j.engappai.2025.112130}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112130}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Malicious detection and trust calculation using residual recurrent neural network for trust with quality of service-aware multicast routing in mobile ad-hoc network system}, volume = {161}, year = {2025}, }
• (2025). Energy management system scheduling optimization based on an improved generative adversarial network deep reinforcement learning algorithm. EAAI, 161, 112129. (www)

  @article{ , abstract = {Households, as electricity consumers, play a critical role in achieving the carbon peak and carbon neutrality goals. The development of smart grids provides technical support for the efficient integration and distribution of renewable energy, gradually extending to household users. This has led to higher demands for the stability of electricity supply to address the growing electricity demand and the uncertainties associated with renewable energy. To address this, this paper proposes an improved generative adversarial network and an enhanced deep reinforcement learning algorithm to improve the scheduling capability of the energy management system. First, we introduce an improved wasserstein generative adversarial network that combines stochastic differential equations and autocorrelation penalty terms with the generator. The experimental results demonstrate that the proposed method can generate high-quality time series data. The generated data were used to train our scheduling model, effectively enhancing its generalization capability. Secondly, We introduced the Minmax mechanism to address Q-value estimation bias by utilizing multiple Q-networks. This mechanism first divides the target Q-values into several groups, selects the maximum value from each group, and then takes the minimum among these maxima as the final target Q-value. We applied this mechanism to improve deep reinforcement learning algorithms based on multi-Q-value evaluation. Comparison experiments show that this improvement significantly enhances the algorithm’s performance, outperforming traditional algorithms in terms of convergence, volatility, and final rewards. The energy management system demonstrates stronger adaptability when handling uncertainties arising from renewable energy variations, ensuring reliable power supply and achieving balanced energy management.}, archive = {J_EAAI}, author = {Weipeng Chao and Yuanbo Shi and Yushuai Li and Meng Liu and Xiaoling Leng}, doi = {10.1016/j.engappai.2025.112129}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112129}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Energy management system scheduling optimization based on an improved generative adversarial network deep reinforcement learning algorithm}, volume = {161}, year = {2025}, }
• (2025). Deep ensemble learning and error correction method for remaining useful life prediction of rolling bearings. EAAI, 161, 112128. (www)

  @article{ , abstract = {Rolling bearings are critical components of rotating machinery, and the prediction of their remaining useful life (RUL) is important for system reliability and operating efficiency. A novel RUL prediction method based on deep ensemble learning and error correction is here proposed. Firstly, the moving average filter (MAF) is applied to preprocess vibration signals for removing outliers and noise. Then, time-domain features are extracted from the processed vibration signals, and optimized to obtain imperative feature signals. Subsequently, a deep ensemble learning model is built with gated recurrent unit (GRU), bidirectional GRU (BiGRU), long short-term memory (LSTM) and bidirectional LSTM (BiLSTM) as base learners, and the overall performance of the prediction model is enhanced by introducing an error correction strategy. The MAF method is also used to smooth the trend of the RUL prediction outcomes. Finally, the proposed method is applied to two full-lifecycle rolling bearing datasets: the Prognostics and Health Management 2012 (PHM2012) dataset and the Intelligent Maintenance System (IMS) dataset. It is evaluated using mean square error (MSE), mean absolute error (MAE), and the R-square coefficient of determination (R 2 ). The test results demonstrate that the method achieves highly accurate RUL predictions: on the PHM2012 dataset, the MSE, MAE, and R 2 reach 0.000380, 0.013695, and 0.994716, respectively; on the IMS bearing dataset, the corresponding values are 0.001056, 0.015403, and 0.978346. In addition, the method outperforms traditional single models (GRU, BiGRU, LSTM, BiLSTM) as well as the Transformer model in both cases, with R 2 improvements over the Transformer of 0.011065 and 0.008744, respectively. These results highlight the superior generalization capability and robustness of the proposed method, making it applicable to industrial environments requiring reliable RUL prediction.}, archive = {J_EAAI}, author = {Wenzhe Yin and Hong Xia and Enrico Zio and Xueying Huang}, doi = {10.1016/j.engappai.2025.112128}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112128}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep ensemble learning and error correction method for remaining useful life prediction of rolling bearings}, volume = {161}, year = {2025}, }
• (2025). Maritime supply chain optimization using robust adversarial reinforcement learning. EAAI, 161, 112127. (www)

  @article{ , abstract = {This study explores novel strategies for analyzing and managing port productivity in a multi-stage supply chain network by integrating synchronization and reinforcement learning (RL) techniques. Current port management systems face issues with nonlinearity, high interdependency, and vulnerability to market disruptions, which might destabilize port operations. To tackle these issues, seaport operations are examined in four stages of implementation: the terminal operator, inland carrier, inland terminal operator, and consignee. Brownian motion is applied to characterize stochastic disruptions in volatile markets, and the port performance under market disruptions is analyzed using nonlinear data analytics. The dynamical analysis reveals that port management systems exhibit highly coupled nonlinear dynamics with a tendency towards instability. The complex nature of maritime port logistics requires innovative strategies to analyze container handling volumes, optimize the strategic planning of port management, and improve overall efficiency. A novel optimal policy for port operations is realized by integrating a deep deterministic policy gradient into robust adversarial deep learning. A deep reinforcement learning algorithm is employed to learn adaptively from historical port-related data and real-time container handling feedback, enabling intelligent strategies to make informed decisions and dynamically adjust policies in response to stochastic disruptions or changing market conditions. Quantitative results demonstrate that the proposed strategy achieves up to 97.78 % operating efficiency despite disturbances, and the adversarial attacks are shown to decrease port productivity by up to 77.78 % in scenarios without robust deep learning support. This study contributes to the growing field of intelligent port operations by paving the way for more adaptive and smart solutions in the maritime supply chain network.}, archive = {J_EAAI}, author = {Truong Ngoc Cuong and Sam-Sang You and Le Ngoc Bao Long and Hwan-Seong Kim and Duy Anh Nguyen and Nguyen Duy Tan}, doi = {10.1016/j.engappai.2025.112127}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112127}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maritime supply chain optimization using robust adversarial reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Employing dual-path structure and soft attention mechanism to enhance recognition and classification of wild medicinal licorice in xinjiang. EAAI, 161, 112126. (www)

  @article{ , abstract = {Licorice is highly valued in traditional Chinese medicine for its anti-inflammatory, antiviral, and immunomodulatory properties, and is widely used in the pharmaceutical, food, and cosmetic industries. Xinjiang, the largest licorice-producing region in China, faces severe overharvesting of wild licorice due to increasing market demand, threatens natural populations and fragile ecosystems. Accurate identification and classification of licorice species are crucial for environmental protection and sustainable resource utilization, as traditional methods relying on experience are inefficient, subjective, and prone to errors. This study builds on the Inception-Residual Network-Version 2 (Inception-ResNet-V2) architecture and proposes an advanced licorice recognition model called Inception-ResNet-V2-Soft Attention, Dual-path Structure, and Focal Loss (IRV2-SDF). The IRV2-SDF model integrates a soft attention mechanism that focuses on key regions, a dual-path structure for multi-scale feature extraction, and a focal loss function to address class imbalance. It aims to improve the identification and classification of three wild licorice species ( Glycyrrhiza glabra , Glycyrrhiza inflata , and Glycyrrhiza uralensis ) and associated weeds in complex environments. Trained on 3,653 images collected from Xinjiang, the model achieves an average recognition accuracy of 91.79%, surpassing traditional models, with accuracy improvements of 4.27%, 2.08%, 2.76%, and 6.36% for G. glabra , G. inflata , G. uralensis , and weeds, respectively. By effectively reducing background noise and enhancing detection capabilities, the model overcomes the limitations of traditional methods and provides a robust solution for wild licorice recognition. This research offers a technical foundation for licorice conservation and sustainable utilization and can serve as a reference for identifying other medicinal plants in complex environments.}, archive = {J_EAAI}, author = {Yuan Qin and Jianguo Dai and Guoshun Zhang and Miaomiao Xu and Jing Yang and Jinglong Liu}, doi = {10.1016/j.engappai.2025.112126}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112126}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Employing dual-path structure and soft attention mechanism to enhance recognition and classification of wild medicinal licorice in xinjiang}, volume = {161}, year = {2025}, }
• (2025). Real-time cooperative target tracking in cluttered environments using multiple drone swarms with adaptive fuzzy emotional learning. EAAI, 161, 112125. (www)

  @article{ , abstract = {This paper presents a real-time adaptive trajectory prediction framework for cooperative unmanned aerial vehicle (UAV) swarms engaged in dynamic target tracking within cluttered environments. The proposed system introduces a novel artificial intelligence (AI)-based control architecture combining fuzzy inference, neuro-emotional learning, and distributed multi-agent coordination. At the core of the approach is a Bidirectional Fuzzy Brain Emotional Learning Prediction (BFBEL-P) model, which integrates fuzzy logic and an online adaptive neural structure to enable trajectory forecasting without pre-training or prior knowledge of the environment. From an engineering perspective, this AI model is deployed in UAV swarm navigation, where robust decision-making, predictive coordination, and obstacle avoidance are essential for target interception missions. In contrast to conventional prediction methods, such as curve fitting, nonlinear model predictive control (MPC), and deep learning-based Long Short-Term Memory (LSTM) networks, the BFBEL-P framework offers fast convergence, low computational cost, and high adaptability. The system incorporates multi-threaded data fusion across the swarm to achieve consensus-driven predictions and maintain situational awareness, even under sensor failures or occlusions. Simulation results show that BFBEL-P improves short-term prediction accuracy by 82.2%, reduces prediction time by 15%, and achieves a 100% tracking success rate across benchmark scenarios. These results establish BFBEL-P as a reliable AI technique for distributed control in real-world UAV swarm applications, offering a promising tool for search-and-rescue, surveillance, and defense operations.}, archive = {J_EAAI}, author = {Lucas William Page and Vu Phi Tran and Duy Luan Nguyen}, doi = {10.1016/j.engappai.2025.112125}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112125}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time cooperative target tracking in cluttered environments using multiple drone swarms with adaptive fuzzy emotional learning}, volume = {161}, year = {2025}, }
• (2025). Hierarchical multi-scale matched masked autoencoder for industrial multi-rate time series modeling. EAAI, 161, 112122. (www)

  @article{ , abstract = {In practical industrial processes, due to sensor hardware limitations, the sampling rates of different variables often vary, leading to multi-rate time series (MRTS) data. However, the distribution of multi-scale dynamics in MRTS data typically follows a step-like pattern, with intricate scale transitions from fine to coarse and complex scale-consistent dependencies across rates. Additionally, the inherent characteristics of MRTS data often result in label scarcity. Both factors present significant challenges for MRTS modeling. To address these issues, we propose a novel self-supervised learning strategy, called H ierarchical M ulti-Scale M atched M asked A utoencoder (H3MAE). Specifically, we design a scale-matching input fusion mechanism where each layer is hierarchically aligned to a specific scale, with the scale-matching integration from two sources, effectively capturing the multi-scale dynamics and cross-rate scale-consistent dependencies in MRTS data. Besides, we introduce a novel auxiliary task that imputes masked positions in the encoded representation space at each layer, aiming to achieve MRTS representation learning and mitigate label scarcity. Furthermore, we propose a unique encoder-imputer structure in each layer to enable multi-scale self-supervised learning while generating temporally aligned features satisfying the input requirements of the next layer. Experimental results on three benchmark datasets and two industrial multi-rate tasks demonstrate that our framework yields better performance in MRTS modeling. The code is publicly available at https://github.com/monolithycq/H3MAE .}, archive = {J_EAAI}, author = {Changqing Yuan and Yongfang Xie and Shiwen Xie and Jie Wang}, doi = {10.1016/j.engappai.2025.112122}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112122}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical multi-scale matched masked autoencoder for industrial multi-rate time series modeling}, volume = {161}, year = {2025}, }
• (2025). Unsupervised feature selection via latent feature representation and modified graph embedding. EAAI, 161, 112121. (www)

  @article{ , abstract = {Data in practical applications are correlated not only between samples but also between high-dimensional features. Latent representations can effectively represent such correlations. Existing latent representation methods mainly consider inter-instance relationships and rely on constructing similarity graphs. However, the interconnection information obtained by latent representations is redundant due to the noise and irrelevance of the raw data. Therefore, to address the above issues, this paper proposes an unsupervised feature selection by latent feature representation and modified graph (LFRMG, for short) embedding. First, a latent feature representation norm for learning self-representation structure is designed to explore the interconnections among features. Mining feature relationships via self-representation can solve the redundancy caused by fixed similarity graph relationships. Second, the latent feature representation is combined with modified graph regularization to achieve good feature interconnections while maintaining local data information. In addition, the l 2 , 0 -norm is utilized to select the full-row sparse projection of salient features to avoid the drawbacks of sparsity limitation and parameter tunability. A scheme for solving the closed form of l 2 , 0 is constructed. Finally, the presented approach is superior to many state-of-the-art unsupervised methods through comprehensive experiments on nine existing datasets.}, archive = {J_EAAI}, author = {Jialing Yan and Gang Hu and Guo Wei}, doi = {10.1016/j.engappai.2025.112121}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112121}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised feature selection via latent feature representation and modified graph embedding}, volume = {161}, year = {2025}, }
• (2025). Coarse-to-fine dual-branch network for ship target recognition in complex environments. EAAI, 161, 112120. (www)

  @article{ , abstract = {Harsh sea conditions and the complex and variable positions of ships significantly impact the capacity of imaging devices to capture high-quality ship images, making ship target recognition challenging in the application of artificial intelligence. Many scholars have recently proposed cascaded recognition models to address this issue. Following this method, in this paper, we propose a novel method for ship target recognition in complex environments called the coarse-to-fine dual-branch (CFDB) network. The CFDB model designs a dual-branch network from coarse to fine to lock the target area fine features and then uses peer-to-peer communication to extract and exchange learning of the target region’s final discriminative contour features, assisting in predicting ship classes in the complex environment. The proposed method is evaluated on the constructed complex in background ships (CIB-ships) dataset and the publicly available Marine Argos Recognition Ships (MAR-ships) and Game-of-Ships datasets. Compared with the suboptimal method, the proposed CFDB network exhibits improvements of 2.11%, 1.33%, and 1.24% accuracy on the CIB-ships, MAR-ships, and Game-of-ships datasets, respectively. The results demonstrate that the proposed method provides useful ideas for the dynamic monitoring of ships in real environments. Our code will be published at https://github.com/yangt1013/CFDB-master .}, archive = {J_EAAI}, author = {Yang Tian and Hao Meng}, doi = {10.1016/j.engappai.2025.112120}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112120}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coarse-to-fine dual-branch network for ship target recognition in complex environments}, volume = {161}, year = {2025}, }
• (2025). A multi-scale sparse channel transformer network for image reconstruction of astronomical bright source contamination. EAAI, 161, 112119. (www)

  @article{ , abstract = {Bright source contamination has long been a challenging issue in the field of image processing, particularly in applications such as astronomical observations, satellite imaging, and nighttime surveillance. To address this issue, this paper proposes a novel Multi-Scale Sparse Channel Transformer Network (MSCformer) aimed at achieving high-quality image reconstruction under the influence of bright source contamination. The network integrates a Top-k Sparse Attention mechanism with a Channel Attention module, enabling selective focus on the most informative features and adaptive weight allocation across channels. Additionally, a Multi-Scale Dual-Gate Feedforward Network is designed to further enhance the expression of valuable features while suppressing redundant information. Experimental results demonstrate that the proposed method exhibits outstanding performance in practical applications on the Sloan Digital Sky Survey (SDSS) photometric image dataset. Compared to existing state-of-the-art techniques, MSCformer achieves significant performance improvements, with a Peak Signal-to-Noise Ratio (PSNR) of 45.093 decibel(dB), a Structural Similarity Index Measure(SSIM) of 0.978, and a Pixel Average Absolute Error (PAAE) of 0.675. This not only significantly enhances the removal of bright source contamination in the field of astronomy but also provides important reference value for subsequent research in related domains.}, archive = {J_EAAI}, author = {Yajuan Zhang and Congcong Shen and Xia Jiang and Bo Qiu and Ali Luo and Fuji Ren and Yuanlu Chen}, doi = {10.1016/j.engappai.2025.112119}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112119}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale sparse channel transformer network for image reconstruction of astronomical bright source contamination}, volume = {161}, year = {2025}, }
• (2025). Seismic damage assessment of reinforced concrete frame structures based on transfer-residual networks. EAAI, 161, 112118. (www)

  @article{ , abstract = {Post-earthquake building damage assessment is a critical research focus in civil engineering due to its vital role in guiding timely and informed rescue operations. This study proposes a rapid seismic damage assessment method for reinforced concrete frame structures based on transfer-residual networks. The methodology involves pre-training a base model on a large dataset generated from computationally efficient simplified model simulations, followed by transfer learning on a smaller, high-fidelity dataset derived from refined finite element model simulations. This approach significantly enhances prediction accuracy while reducing computational costs for seismic damage assessment in new building structures. Optimization of model parameters was performed to find the optimal residual network, which achieves an accuracy of 91.2 % with a remarkable 32-fold speedup over conventional nonlinear time-history analysis methods. Moreover, the transfer-residual network enhances accuracy by 10 % compared to training from scratch. The results conclusively demonstrate the feasibility of utilizing large datasets from simplified models to improve the training accuracy of refined models with limited data, providing a valuable reference for advancing transfer learning-based frameworks in seismic damage assessment of buildings.}, archive = {J_EAAI}, author = {Chen Xiong and Zhijie Luo and Jie Zheng and Linlin Xie and Liu Mei and Lixiao Li and Wujian Long}, doi = {10.1016/j.engappai.2025.112118}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112118}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seismic damage assessment of reinforced concrete frame structures based on transfer-residual networks}, volume = {161}, year = {2025}, }
• (2025). Machine learning-enabled performance-based design of three-dimensional printed engineered cementitious composites. EAAI, 161, 112117. (www)

  @article{ , abstract = {The superior tensile ductility of engineered cementitious composites (ECC) offers a promising solution to the challenge of integrating conventional steel reinforcement in three-dimensional (3D) concrete printing (3DCP). However, the widespread adoption of 3D printed ECC (3DP-ECC) is hindered by the reliance on trial-and-error design process. The complex material component and inherent anisotropy of 3DP-ECC pose challenges for accurate property prediction and inverse design. This paper introduces a performance-based design strategy for 3DP-ECC, leveraging machine learning (ML) and multi-objective optimization. The anisotropic-mechanical properties including compressive strength and flexural strength were experimentally and statistically investigated; further, ML prediction models conbined with multi-objective optimization algorithm were developed to inversely design 3DP-ECC for specific mechanical performance requirements, while reducing carbon footprint and material cost. Specifically, an extensive database was assembled, followed by grey relational analysis (GRA) to identify the parametric sensitivity of the mechanical properties of 3DP-ECC. Three representative ML techniques were employed, with the back-propagation artificial neural network (BPANN) demonstrating superior predictive accuracy. Model interpretability analyses uncovered the importance of input parameters and their influence on predicted outcomes. Lastly, non-dominated Sorting Genetic Algorithm II (NSGA-II) integrated with the BPANN models was applied to perform the inverse design of 3DP-ECC, showing good effectiveness and accuracy. This work offers an efficient and viable avenue for performance-based design for 3DP-ECC, along with the potential to develop low-carbon cost-effective 3DP-ECC.}, archive = {J_EAAI}, author = {Wenguang Chen and Long Liang and Junhong Ye and Lingfei Liu and Neven Ukrainczyk and Liqiang Yin and Jiangtao Yu and Kequan Yu}, doi = {10.1016/j.engappai.2025.112117}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112117}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-enabled performance-based design of three-dimensional printed engineered cementitious composites}, volume = {161}, year = {2025}, }
• (2025). Personalized text-to-image generation with large language and vision assistant enhanced training. EAAI, 161, 112116. (www)

  @article{ , abstract = {Personalized image generation aims to synthesize images of a specific identity. The identity, denoted as V ∗ , refers to an entity with distinctive visual attributes, such as a dog-shaped backpack. However, existing methods like DreamBooth and Custom Diffusion often struggle to generate images of V ∗ that accurately match the input prompts. In this work, we analyze two key issues underlying this limitation: (1) the overbinding problem , where the prompt tokens used to represent V ∗ unintentionally bind to irrelevant visual details from the reference image during training; and (2) the low language prior problem , where insufficient use of pre-trained language prior limits the model’s ability to faithfully generate all the prompt words. To overcome these challenges, we propose LLaVA-Booth, a novel personalization method for diverse, identity-preserving image generation, based on Large Language and Vision Assistant (LLaVA) enhanced training. Our method alleviates the overbinding problem by disentangling background information and solves the low language prior problem by enriching the language context. Additionally, we introduce two auxiliary objectives: (1) an identity (ID) binding loss to strengthen the identity binding and (2) a prior preservation loss to prevent language drift and encourage generation diversity. Experiments demonstrate that LLaVA-Booth effectively mitigates overbinding and enhances language priors to improve prompt fidelity, then generates diverse, high-quality, and identity-preserving images of V ∗ .}, archive = {J_EAAI}, author = {Junsheng Luan and Zhanjie Zhang and Wei Xing and Lei Zhao}, doi = {10.1016/j.engappai.2025.112116}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112116}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalized text-to-image generation with large language and vision assistant enhanced training}, volume = {161}, year = {2025}, }
• (2025). A multi-view new energy vehicle form generation design method combining kansei imagery and deep learning. EAAI, 161, 112115. (www)

  @article{ , abstract = {In the competitive landscape of new energy vehicles, exterior design has become a crucial differentiator amid functional homogenization. User preferences are central to shaping vehicle appearance, yet most perceptual design methods rely on a single viewpoint, limiting insights into complex preference patterns. This study proposes a multi-perspective mapping approach that integrates Kansei engineering with deep learning. Firstly, user core imagery is collected and mined through big data. Secondly, Kernels Network (KNet) semantic segmentation model, Residual Networks (ResNet) tri-view (front/side/rear) score prediction model and fully connected network (FCN) feature fusion model are integrated to construct a multi-view feature mapping system. Finally, the optimal combination of morphological elements is explored based on the Elite Genetic Algorithm (EGA), and the scheme is validated through generative artificial intelligence (AI) workflow. The experimental results demonstrate that, employing “Cool” as a case study, the three-view scheme and the combination scheme devised by this research process exhibit substantial superiority over the majority of the samples. Under identical parameters, the scheme with decision constraints surpasses the randomly generated scheme in terms of perceptual scores and stability. The performance of the test set and the experimental results collectively substantiate the model’s validity. This workflow—covering preference extraction, morphological decomposition, AI-driven generation, and validation—provides a scalable framework for new energy vehicle exterior design. It also demonstrates novel applications of Kansei engineering in multi-view fusion and generative form design.}, archive = {J_EAAI}, author = {Zihao Wang and Le Xi and Yifan Ding and Wenjie Fang and Kaiming Wang and Hongliang Zuo}, doi = {10.1016/j.engappai.2025.112115}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112115}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-view new energy vehicle form generation design method combining kansei imagery and deep learning}, volume = {161}, year = {2025}, }
• (2025). Explainable remaining useful life uncertainty prediction method for rolling bearing. EAAI, 161, 112114. (www)

  @article{ , abstract = {Rolling bearing remaining useful life prediction is the core technology of equipment maintenance. Although deep learning-based prediction methods have made significant breakthroughs, the problems of insufficient model explainability and prediction result uncertainty quantification have seriously constrained the credibility of maintenance decisions. Therefore, this research combines prediction uncertainty quantification with model explanation to propose an explainable uncertainty prediction method. The method includes multi-dimensional feature extraction, remaining useful life uncertainty prediction, and Shapley additive explanations interpreter. For feature extraction, multi-dimensional feature vectors are constructed as network inputs by extracting time-domain features and frequency-domain features. Then, the remaining useful life prediction interval for the rolling bearing is compressed by the proposed gated temporal quantile network. Finally, the prediction model is explained using the Shapley additive explanations interpreter. The multi-case validation results based on the Xi'an Jiaotong University and Changxing Sumyoung Technology Co., Ltd. (XJTU-SY) and Intelligent Maintenance Systems (IMS) rolling bearing full life cycle datasets show that the proposed model has an average interval coverage of 93.96 % and an average interval width of 9.92 %, which indicates that the model maintains high accuracy and robustness in different cases. The nonlinear mapping relationship between the prediction results and the features is clarified by analyzing the Shapley values. Finally, the Shapley values are used to rank the importance of the features to locate the position that may cause rolling bearing performance degradation, which provides credible decision support for the development of the predictive maintenance strategy.}, archive = {J_EAAI}, author = {Ting Zhang and Honglei Wang}, doi = {10.1016/j.engappai.2025.112114}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112114}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable remaining useful life uncertainty prediction method for rolling bearing}, volume = {161}, year = {2025}, }
• (2025). Machine learning-based estimation of sound transmission loss in single and double-layered rectangular functionally graded plates. EAAI, 161, 112112. (www)

  @article{ , abstract = {This study presents a robust machine learning (ML) framework for predicting sound transmission loss (STL) in temperature-dependent functionally graded (FG) single- and double-layered plates, combining analytical modeling and data-driven approaches. The FG plates are modeled with material properties varying along the thickness via a power-law distribution, incorporating nonlinear temperature effects. A high-fidelity analytical solution is derived using first-order shear deformation theory (FSDT) and acoustic velocity potential, with governing equations formulated via Hamilton's principle and solved using the weighted residual method. To enable ML training, a large-scale parametric study generates 189,000 data points, covering variations in plate geometry, temperature, acoustic cavity depth, gradient index, and incident wave angles. ML algorithms are trained on this data to predict STL, with the extreme gradient boosting (XGBoost) algorithm demonstrating the highest accuracy (coefficient of determination R 2 = 99.38 % for training data and R 2 = 99.11 % for test data). The validated ML model is then employed to investigate key parameters—temperature, power-law index, incidence angle, cavity depth, and plate thickness—revealing their nonlinear interactions and impact on STL performance. The developed artificial intelligence (AI) framework provides an efficient tool for acoustic design optimization, offering real-time parameter tuning capabilities for engineers working with functionally graded acoustic barriers.}, archive = {J_EAAI}, author = {Chunfeng Jiang and Masoud Babaei}, doi = {10.1016/j.engappai.2025.112112}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112112}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based estimation of sound transmission loss in single and double-layered rectangular functionally graded plates}, volume = {161}, year = {2025}, }
• (2025). Cubic linguistic T-spherical fuzzy aggregation operator-based multi-attribute group decision-making model and its application to food waste treatment technique selection. EAAI, 161, 112111. (www)

  @article{ , abstract = {Food waste presents significant environmental, economic, and social challenges globally, contributing to resource depletion and greenhouse gas emissions. Effective treatment techniques are essential for minimizing these impacts, promoting sustainability, and supporting circular economy practices. However, selecting the most suitable food waste treatment technique is a complex multi-attribute group decision-making (MAGDM) problem that involves the simultaneous consideration of environmental, economic, and social factors under significant uncertainty. To address this challenge, this study proposes a novel cubic linguistic T-spherical fuzzy sets (CLT-SFSs) framework, which integrates linguistic T-spherical fuzzy sets and linguistic interval-valued T-spherical fuzzy sets to enhance decision-making precision under uncertainty. The proposed CLT-SF framework offers a unified structure capable of capturing both qualitative expert opinions and quantitative uncertainty ranges, offering unprecedented expressiveness for complex sustainability assessments. First, the formal definition and basic operations for CLT-SF numbers, including addition, multiplication, scalar multiplication, and scalar power, and a comparison law are established. Next, to efficiently aggregate cubic linguistic T-spherical fuzzy information, we propose the cubic linguistic T-spherical fuzzy weighted averaging and the cubic linguistic T-spherical fuzzy weighted geometric aggregation operators. These aggregation operators can effectively and comprehensively aggregate attribute values in MAGDM problems. Subsequently, utilizing these operators, a CLT-SFS-based combinative distance-based assessment model is developed to address MAGDM problems. The model's applicability and robustness are demonstrated through a real-world case study on the selection of food waste treatment techniques. A parameter analysis is also conducted to examine the sensitivity of ranking outcomes. Finally, a comparative analysis with existing methods confirms the proposed model's effectiveness, feasibility, and advantages in addressing complex decision-making scenarios. This research not only advances the theoretical framework of fuzzy decision-making but also provides a practical tool for stakeholders in waste management to make informed and sustainable choices.}, archive = {J_EAAI}, author = {Shahid Hussain Gurmani and Weiping Ding and Rana Muhammad Zulqarnain and Jiangfeng Hao}, doi = {10.1016/j.engappai.2025.112111}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112111}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cubic linguistic T-spherical fuzzy aggregation operator-based multi-attribute group decision-making model and its application to food waste treatment technique selection}, volume = {161}, year = {2025}, }
• (2025). Contextual and orientation correction modules enhance weakly-supervised aerial object detection in remote sensing images. EAAI, 161, 112110. (www)

  @article{ , abstract = {This paper addresses the challenges of context and orientation ambiguity in weakly supervised aerial object detection. While current research focuses on improving detection accuracy and efficiency, it often encounters difficulties with contextual and rotational variations in aerial imagery. We propose a novel Context and Orientation Correction (COC) framework, which includes two innovative modules: a context correction module and an orientation correction module. The context correction module utilizes style normalization to guide the model in identifying atypical objects within specific contextual scenes by mitigating contextual disparities between instances and refining contextual information. Additionally, the orientation correction module aims to reduce feature distance between instances with varying orientations, leveraging contrastive learning to ensure consistent object representations. Furthermore, we introduce a category-aware aggregation loss to enhance similarity in feature representations of objects from the same category, thereby addressing the class collision issue commonly associated with contrastive learning. Our COC framework achieves 27.6% mean Average Precision and 59.8% mean Average Precision on the Detection in Optical Remote Sensing Image (DIOR) and Northwestern Polytechnical University Very High Resolution 10. v2 (NWPU VHR-10.v2) datasets, respectively, demonstrating its significant effectiveness.}, archive = {J_EAAI}, author = {Le Yang and Shunzhou Wang and Xuerong Wang and Shutong Wang and Yuting Lu and Binglu Wang}, doi = {10.1016/j.engappai.2025.112110}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112110}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contextual and orientation correction modules enhance weakly-supervised aerial object detection in remote sensing images}, volume = {161}, year = {2025}, }
• (2025). Towards detection of anomalous cosmic ray signals for observations acquired from cosmic ray extremely distributed observatory mobile detectors. EAAI, 161, 112109. (www)

  @article{ , abstract = {In this work, we propose and test a method that allows the detection of anomalous cosmic ray signals acquired using Complementary Metal-Oxide-Semiconductor detectors. The method uses unsupervised embedding based on Principal Component Analysis which we named Eigenhits in apparent analogy to Eigenfaces. The embedding generated using Eigenhits allows the detection of potential anomalies, defined as images whose position described by the embedding relative to a given measure is above a certain distance threshold from other images. Thus, the problem of anomaly detection was reduced to the problem of detecting outliers which can be solved, for example, using clustering algorithms. We conducted tests of our approach on the Cosmic Ray Extremely Distributed Observatory (CREDO) dataset containing 13168 images and obtained satisfactory results demonstrating the stability and effectiveness of the method. The embedding generation method we propose in this paper and the evaluation of its effectiveness in detecting anomalies in images of cosmic ray events from CREDO dataset is a pioneering study with many critical applications.}, archive = {J_EAAI}, author = {Tomasz Hachaj and Łukasz Bibrzycki and Marcin Piekarczyk and Olaf Bar and Michał Niedźwiecki and Sławomir Stuglik and Piotr Homola and Dmitriy Beznosko and David Alvarez-Castillo and Bożena Poncyljusz and Ophir Ruimi and Oleksandr Sushchov and Krzysztof Rzecki and CREDO collaboration}, doi = {10.1016/j.engappai.2025.112109}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112109}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards detection of anomalous cosmic ray signals for observations acquired from cosmic ray extremely distributed observatory mobile detectors}, volume = {161}, year = {2025}, }
• (2025). An uncertainty-aware safe-evolving reinforcement learning algorithm for decision-making and control in highway autonomous driving. EAAI, 161, 112108. (www)

  @article{ , abstract = {Rule-based and optimization-based approaches face challenges in decision-making and control for autonomous vehicles (AVs) in dynamic and complex highway scenarios. In contrast, reinforcement learning (RL) offers a more flexible and adaptable data-driven solution by allowing AVs to learn optimal actions through interactions with the environment, without requiring predefined rules or explicit programming. However, in real-world highway environments characterized by uncertainty, RL algorithm encounter difficulties in ensuring stability and safety. To address these challenges, this paper proposes an uncertainty-aware safe-evolving RL algorithm that integrates internal stability, external stability, and provable safety mechanisms. The internal stability mechanism ensures consistent performance improvements with high probability during policy updates itself, while the external stability leverages a benchmark policy as a reference to ensure the current policy performs at least as well as, if not better than, the benchmark. Furthermore, an action projection mechanism and a mixed learning procedure are incorporated to make minimal modifications to the learned policy, ensuring safety while supporting stable learning from both safe and original actions. The results show that the proposed algorithm maintains stability and safety throughout the learning process, achieves final performance comparable to traditional RL methods, and delivers higher training efficiency in a complex dynamic highway scenario in simulation. This suggests that the algorithm offers a viable solution for self-evolving systems in uncertain real-world environments, where traditional approaches may struggle.}, archive = {J_EAAI}, author = {Ping Lu and Sunan Zhang and Feihong Tan and Fulin Zhang and Yuxiang Feng and Bo Hu}, doi = {10.1016/j.engappai.2025.112108}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112108}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An uncertainty-aware safe-evolving reinforcement learning algorithm for decision-making and control in highway autonomous driving}, volume = {161}, year = {2025}, }
• (2025). A deep learning-driven network for joint low-light enhancement and deblurring in maritime surveillance systems. EAAI, 161, 112107. (www)

  @article{ , abstract = {—Shipboard video surveillance systems are crucial for maritime environmental perception. However, shipborne cameras are often affected by challenges such as ship rolling and low illumination, causing motion blur and low-light degradation in captured images. These degradations significantly reduce the accuracy and robustness of sea-surface obstacle detection. To address this challenge, this paper presents a deep learning-based joint network model designed to simultaneously deblur and enhance low-light maritime images. Unlike existing step-by-step approaches that treat deblurring and low-light enhancement as separate tasks, the proposed model employs an encoder-decoder architecture to jointly address both tasks within a unified framework, thereby overcoming the limitations of separate processing. Furthermore, the model incorporates a novel adaptive wavelet curve attention mechanism and an extended spatial pyramid pooling module with dynamic global context to mitigate detail loss caused by sea surface reflections and artifacts in dynamic backgrounds. Additionally, a joint loss function is exploited to further enhance image restoration quality. Experiments on a self-constructed synthetic maritime dataset demonstrate that the proposed model achieves a Peak Signal-to-Noise Ratio (PSNR) of 31.81, a Structural Similarity Index Measure (SSIM) of 0.812, and a Learned Perceptual Image Patch Similarity (LPIPS) of 0.246, significantly outperforming existing competing methods. Moreover, evaluations on real-world degraded maritime images confirm the model's strong generalization ability and robustness. This advancement offers practical benefits for improving obstacle perception in autonomous ships navigating complex maritime environments.}, archive = {J_EAAI}, author = {Zeyuan Shao and Yong Yin and Hongguang Lyu and Qianfeng Jing and Tao Cheng}, doi = {10.1016/j.engappai.2025.112107}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112107}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning-driven network for joint low-light enhancement and deblurring in maritime surveillance systems}, volume = {161}, year = {2025}, }
• (2025). A double hierarchy hesitant fuzzy forecasting model considering the influence of investor emotion and the co-movement of stock markets. EAAI, 161, 112106. (www)

  @article{ , abstract = {The stock market is a complex system influenced by various internal and external factors. Effective stock index forecasting models must balance interpretability with the ability to integrate multi-source information. While traditional fuzzy models offer high interpretability, they struggle to incorporate multiple dynamic factors. This study proposes a novel forecasting model based on double hierarchy hesitant fuzzy linguistic term sets (DHHFLTS), which integrates investor sentiment and international market co-movement. By constructing fuzzy logic rules, the model captures complex interactions underlying index fluctuations and enhances predictive accuracy in volatile environments. To evaluate its effectiveness, the model was tested on time series data from three major indices: the Shanghai Stock Exchange Composite Index (SSEC), Taiwan Capitalization Weighted Stock Index (TAIEX), and Financial Times Stock Exchange 100 Index (FTSE 100). The Root Mean Square Error (RMSE) achieved was 32.06 for the SSEC, 69.74 for the TAIEX, and 25.27 for the FTSE 100. Compared with existing benchmark methods, the proposed model demonstrates superior predictive accuracy and generalization performance. These findings confirm the model's capability in capturing multidimensional uncertainty and temporal patterns, highlighting its potential for intelligent financial forecasting and decision-making.}, archive = {J_EAAI}, author = {Liu Zhengmin and Du Chuantao and Zhang Jihao and Gao Shanshan and Liu Peide}, doi = {10.1016/j.engappai.2025.112106}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112106}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A double hierarchy hesitant fuzzy forecasting model considering the influence of investor emotion and the co-movement of stock markets}, volume = {161}, year = {2025}, }
• (2025). Style prompt tuning for bridging visual gaps in autonomous driving. EAAI, 161, 112105. (www)

  @article{ , abstract = {Artificial intelligence models for semantic segmentation and image classification in autonomous driving must maintain reliability across adverse conditions such as rain, fog, snow, and nighttime. However, models trained on only clear daytime images often fail to generalize under such domain shifts. Existing unsupervised domain adaptation (UDA) methods employ image-level style transfer using generative adversarial networks (GANs) or diffusion models, which necessitate paired data and risk altering content. Therefore, this study proposes Style Prompt Tuning, a novel UDA framework that utilizes image–text models to automatically generate and optimize textual prompts representing target-domain styles. These prompts guide a U-Net-based style network to synthesize source images in the target style while preserving their semantic content. Our approach employs clustering within the Contrastive Language—Image Pretraining (CLIP) embedding space and a composite loss function, including content, style, transfer, patch, and total variation terms to optimize prompt quality. The generated stylized images augment the source dataset and are used to train more robust task models. Experiments on semantic segmentation benchmarks (Cityscapes-to-Adverse Conditions Dataset with Correspondences (ACDC), DarkZurich, BDD100k-night, and Nighttime Driving) and image classification (Visual Domain Adaptation 2017) reveal our approach to achieve improvements of +3.4 mean intersection-over-union (mIoU) and +0.8% accuracy over prior UDA methods. These results highlight our method’s practical effectiveness for real-world autonomous driving applications under visually challenging scenarios.}, archive = {J_EAAI}, author = {Suyeon Cha and Giyun Choi and Minji Kwak and Jongwon Choi}, doi = {10.1016/j.engappai.2025.112105}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112105}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Style prompt tuning for bridging visual gaps in autonomous driving}, volume = {161}, year = {2025}, }
• (2025). Enhancing fuzzy cognitive map convergence through supervised and unsupervised learning algorithms: A case study of operational risk assessment in power distribution networks. EAAI, 161, 112104. (www)

  @article{ , abstract = {Fuzzy Cognitive Maps (FCMs) are commonly used for modeling complex systems. However, their convergence challenges significantly limit their accuracy and reliability, especially in operational risk assessment. To address this issue, the current study proposes a novel approach that integrates advanced supervised and unsupervised learning algorithms: specifically, the Mesh Adaptive Direct Search (MADS) and Genetic Algorithm (GA). To meet the critical need for accurate risk modeling in power distribution networks, the proposed methodology utilizes ten years of time-series data from the Yazd Power Distribution Network as a real-life case study. This optimizes risk relationships and reduces convergence errors. The main contributions of this research are: (1) the development of an integrated FCM-based framework that improves convergence stability and accuracy through advanced learning algorithms, (2) the demonstration of the effectiveness of MADS in achieving faster convergence and lower error rates compared to GA, and (3) the provision of a data-driven, scalable solution for risk prioritization and decision-making in complex and dynamic systems. The results show a significant decrease in convergence error from 0.126 to 0.005, allowing for more precise risk analysis and mitigation strategies.}, archive = {J_EAAI}, author = {Elham Fallah Baghemoortini and Davood Shishebori and Mustafa Jahangoshai Rezaee and Armin Jabbarzadeh}, doi = {10.1016/j.engappai.2025.112104}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112104}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing fuzzy cognitive map convergence through supervised and unsupervised learning algorithms: A case study of operational risk assessment in power distribution networks}, volume = {161}, year = {2025}, }
• (2025). Semi-supervised generative adversarial network for plant leaf disease detection. EAAI, 161, 112103. (www)

  @article{ , abstract = {Fully supervised plant leaf disease segmentation methods based on convolutional neural networks (CNN) require large quantities of labeled images for training, which are time-consuming and labor-intensive to obtain in practical scenarios. Moreover, most diseases of plant leaf exhibit indistinct edge information and background noise interference, which hinders precise disease localization. To overcome these issues, a semi-supervised generative adversarial network (SSGAN) is proposed for plant leaf disease inspection. Firstly, to learn comprehensive semantic features from limited data, a semi-supervised method based on generative adversarial networks is developed to enhance the interaction of features among labeled and unlabeled disease images. Secondly, a boundary feature attention module (BFAM) is proposed to enhance edge detail representation, which makes the model pay more attention to the boundary feature of plant leaf diseases. Finally, a background noise suppression module (BNSM) is proposed to bolster the differentiation between the normal and disease areas so as to alleviate the adverse impact of background noise. The effectiveness of SSGAN is verified on two plant leaf disease datasets. The mean intersection over union (mIoU) on the apple and tomato leaf datasets improves by 2.07 % and 2.28 % respectively, compared with the suboptimal method. The testing results of experiments show that SSGAN achieves great performance on plant leaf disease segmentation under limited labeled data.}, archive = {J_EAAI}, author = {Lixiang Zhao and Jun Hao and Demin Li and Jianbo Yu}, doi = {10.1016/j.engappai.2025.112103}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112103}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised generative adversarial network for plant leaf disease detection}, volume = {161}, year = {2025}, }
• (2025). Industrial device-aided data collection for real-time rail defect detection via a lightweight network. EAAI, 161, 112102. (www)

  @article{ , abstract = {Rail defect detection is challenging due to the diverse and irregular nature of defects, along with the limited availability of high-quality datasets. Existing methods struggle with effectively capturing multi-scale features for proper feature allocation and preserving crucial details in deep networks, leading to incomplete defect representation and reduced accuracy. To address these limitations, we propose Rail Defect Detection Network (REDNet), a lightweight deep learning model specifically designed for real-time rail defect detection in manufacturing and maintenance applications. We design the Multi-Scale Deep Feature Aggregation (MSDFA) module to enhance semantic consistency modeling and achieve more precise feature fusion. We develop the Adaptive Task Decomposition Head (ATDH) to address dynamic feature allocation, and we introduce the Reversible Column Network (RevCol) as the backbone to enhance feature extraction and ensure information reconstruction. Additionally, we developed a high-quality dataset using specialized equipment to address data scarcity and utilized it for training. REDNet achieved a high mean Average Precision at an Intersection over Union (IoU) threshold of 0.50 (mAP50) of 94.1% with exceptional real-time performance at 204.1 frames per second (FPS), while keeping an efficient design of 5.70 million parameters and surpassing state-of-the-art (SOTA) methods in accuracy and speed. These features make it suitable for defect detection, facilitating engineering deployment, and improve quality control in rail manufacturing and maintenance. Generalization tests on the public Microsoft Common Objects in Context (MS COCO) dataset yielded a mean Average Precision across IoU thresholds from 0.50 to 0.95 (mAP50–95) of 46.8%, further confirming the effectiveness of REDNet.}, archive = {J_EAAI}, author = {Qing Dong and Tianxin Han and Gang Wu and Lina Sun and Min Huang and Fu Zhang}, doi = {10.1016/j.engappai.2025.112102}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112102}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Industrial device-aided data collection for real-time rail defect detection via a lightweight network}, volume = {161}, year = {2025}, }
• (2025). A quantitative multi-dimensional resilience framework in process industries. EAAI, 161, 112100. (www)

  @article{ , abstract = {Process industries face significant safety challenges arising from the handling of hazardous materials, tightly coupled operations, and complex system interdependencies. Addressing these challenges requires management strategies that are both proactive and holistic rather than purely reactive. Resilience engineering embodies this approach by emphasizing a system's ability to anticipate disturbances, adapt under stress, and recover effectively. This study introduces a comprehensive, multi-dimensional framework for assessing resilience in process industries. The framework was developed through content analysis of expert interviews, a systematic literature review, and application of the spherical fuzzy Delphi method. Results revealed five primary dimensions of resilience: the organizational dimension, which encompasses leadership commitment, safety culture, and communication processes; the human resources dimension, focusing on workforce competence, training, and well-being; the individual dimension, addressing cognitive readiness, situational awareness, and decision-making skills; the risk management system dimension, which includes hazard identification, risk assessment, emergency preparedness, and continuous monitoring; and the technical dimension, covering equipment reliability, system redundancy, automation safeguards, and maintenance practices. Within these dimensions, factors such as leadership, effective communication, workforce capability, robust risk governance, and technical robustness emerged as particularly influential. By integrating these dimensions into a unified framework, our study advances theoretical understanding of resilience engineering and offers practical guidance for enhancing safety performance in process plants. Furthermore, this framework lays a foundation for future research aimed at developing standardized assessment tools, evaluating long-term outcomes of resilience interventions, and exploring the integration of emerging technologies—such as artificial intelligence, the internet of things, and advanced automation—into resilience engineering practices.}, archive = {J_EAAI}, author = {Mojtaba Emkani and Moslem Alimohammadlou and Esmaeil Zarei and Mehdi Jahangiri and Mojtaba Kamalinia}, doi = {10.1016/j.engappai.2025.112100}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112100}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A quantitative multi-dimensional resilience framework in process industries}, volume = {161}, year = {2025}, }
• (2025). Mutual transfer learning for cuff-less blood pressure estimation using photoplethysmography-based visibility graphs. EAAI, 161, 112099. (www)

  @article{ , abstract = {Cuff-less blood pressure (BP) estimation is critical to cardiovascular disease prevention and management. Photoplethysmography (PPG)-based monitoring technology offers advantages over cuff-based devices, including portability, lower power consumption, and faster measurements. However, current deep learning methods for BP estimation from PPG signals are limited by their analysis only from one-dimensional perspectives, failing to exploration of the underlying physiological patterns of cross-domain visual representations based on higher dimensional perspectives. Furthermore, conventional knowledge distillation techniques necessitate unidirectional knowledge transfer from pre-trained models, rendering it challenging to obtain feedback on the learning state of small networks for optimizing and adjusting the training process. This is inadequate for acquiring a profound understanding of the intricate mapping relationship between PPG and BP values. Therefore, this work presents a novel Transformer-based mutual transfer learning framework (MTL) that estimates BP values from phase-space reconstructed PPG signals using a multi-field complementary approach. The proposed MTL method leverages four phase-space reconstruction techniques to convert PPG signals into visibility graphs (VGs) that provide rich time-variant information. Furthermore, the joint optimization strategy with multiple losses of the soft label and structural knowledge learning enables us to transfer pre-trained knowledge from heterogeneous Transformer models and obtain cumulative multi-field complementary VG features during the fine-tuning process. We evaluate our MTL on three datasets of 1375 subjects using a subject-wise data-splitting paradigm based on five-fold cross-validation, achieving a state-of-the-art performance with estimation errors of 0.50 ± 4.94 millimeter of mercury (mmHg) and 0.21 ± 2.63 mmHg for systolic and diastolic BP, respectively. Our proposed end-to-end MTL offers a computationally efficient solution and elegant generalization ability for BP estimation using PPG-based VGs, providing a novel and innovative approach to BP monitoring that can advance cardiovascular disease prevention and management.}, archive = {J_EAAI}, author = {Chenbin Ma and Zhenchang Liu and Peng Zhang and Lishuang Guo and Haonan Zhang and Zeyu Liu and Guanglei Zhang}, doi = {10.1016/j.engappai.2025.112099}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112099}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mutual transfer learning for cuff-less blood pressure estimation using photoplethysmography-based visibility graphs}, volume = {161}, year = {2025}, }
• (2025). Groundwater parameters estimation: A hybrid approach of convolutional layers with asynchronous and distributed bio-inspired algorithms. EAAI, 161, 112098. (www)

  @article{ , abstract = {This research focuses on aquifer hydraulic parameters estimation using bio-inspired algorithms since they can tackle groundwater model non-linearities. We propose two novel hybrid frameworks that combine the advantages of convolutional layers (CL) to enhance pattern recognition with heuristic search of Particle Swarm Optimization (PSO) and Differential Evolution (DE) algorithms. These integrations are implemented using an asynchronous and distributed approach to address efficiency issues in large-scale problems, resulting in ADPSO-CL (Asynchronous and Distributed Particle Swarm Optimization with Convolutional Layers) and ADDE-CL (Asynchronous and Distributed Differential Evolution with Convolutional Layers). The distributed method employs virtual machines, where a server generates and assigns particles to workers, which run in parallel with asynchronous iterative solution exchanges. We assess different algorithm configurations in an integrated water management model by coupling two software: Water Evaluation and Planning (WEAP) and MODFLOW. Results indicate that ADPSO-CL outperforms ADDE-CL by demonstrating more stable asynchronous communication, with fewer incomplete experiments (more than one worker was disconnected before completing all iterations), 33% in contrast to 71%. Additionally, produces results closer to the expected values, with mean absolute percentage error (MAPE) values of 78.25% for hydraulic conductivity ( K ) and 55.56% for specific yield ( S y ), compared to 299% and 209% in ADDE-CL. Moreover, ADPSO-CL has the fastest convergence rate, achieving efficient results in about half of the total iterations. This study introduces a novel and scalable architecture for intricate simulation–optimization problems, demonstrating its potential for future applications in real-world water resources planning and management.}, archive = {J_EAAI}, author = {Kiara Tesen and Hermilo Cortés and Sebastián Vicuña and Edmundo Molina-Perez and Francisco Suárez}, doi = {10.1016/j.engappai.2025.112098}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112098}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Groundwater parameters estimation: A hybrid approach of convolutional layers with asynchronous and distributed bio-inspired algorithms}, volume = {161}, year = {2025}, }
• (2025). Forecasting power generation: A novel two-dimensional logistic fractional grey model. EAAI, 161, 112097. (www)

  @article{ , abstract = {Accurate prediction of power generation is essential for the stable operation of power systems, sustained economic growth, and harmonious social development. However, power systems exhibit chaotic characteristics such as complex and nonlinear dynamics during operation; for this reason, this study exploits the fact that two-dimensional logistic mapping helps with the recognition and control of the complex high-dimensional dynamics of the system. Additionally, the fractional-order chaotic system offers a more universal means to describe the chaotic phenomena, etc., and introduces the modelling mechanism of the two-dimensional logistic model and the fractional-order derivative theory into the grey model. A novel two-dimensional logistic fractional-order grey model is developed. The new model can simultaneously analyze two parallel major factors that evolve and are interrelated, and the fractional-order derivatives and nonlinear terms accurately portray the historical state and dynamic evolution process of the system, which can explore the evolution law in the time series of power generation more adequately. Furthermore, the particle swarm optimization algorithm is used to optimize the fractional order derivatives and parameters so that the adjustable model parameters can better capture the change pattern of the original power generation system data. The new model is applied to China's power generation prediction, and its validity is verified from three different perspectives through two different power generation modes as the research objects. Moreover, the new model outperforms the other seven models in its calculations. Finally, the new model effectively forecasts China's power generation for the next five years.}, archive = {J_EAAI}, author = {Mingyue Weng and Huiming Duan and Derong Xie}, doi = {10.1016/j.engappai.2025.112097}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112097}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting power generation: A novel two-dimensional logistic fractional grey model}, volume = {161}, year = {2025}, }
• (2025). A deep reinforcement learning optimization algorithm based on heterogeneous graph neural network for hybrid flow shop scheduling problem with finite transportation resources. EAAI, 161, 112096. (www)

  @article{ , abstract = {In many actual process industries, transportation equipment represented by automated guided vehicles (AGVs) is widely used. However, current related research rarely considers the coordinated scheduling optimization of AGVs and production equipment. To this end, the hybrid flow shop scheduling problem (HFSP) with finite transportation resources (HFSP-FTR) in process industries is considered in this paper. Since it is necessary to consider the coordinated scheduling optimization of machines and AGVs at the same time, an end-to-end deep reinforcement learning scheduling method based on a heterogeneous graph neural network is proposed. First, a heterogeneous graph model capable of expressing any HFSP-FTR instance is constructed to visually represent the allocation relationship among jobs, machines, and AGVs. Second, a Markov decision process is formulated for solving HFSP-FTR, which specifically encompasses state features based on the heterogeneous graph model and a novel representation method for composite scheduling actions. Then, a novel heterogeneous graph neural network framework that can parallelly embed nodes of various types is proposed, aiming to derive graph-level embeddings for HFSP-FTR instances. Finally, a policy network is designed to obtain the probability distribution of each composite scheduling action being executed, and a deep reinforcement learning framework based on the multi-actor network is proposed for training. The results of numerical simulation experiments conducted on test instances with varying types demonstrate that the proposed method can acquire effective and highly generalized scheduling strategies in terms of makespan performance, outperforming other scheduling algorithms that are widely applied regarding the quality of scheduling solutions and other aspects.}, archive = {J_EAAI}, author = {Yejian Zhao and Xiaochuan Luo and Weixiang Xu and Yulin Zhang}, doi = {10.1016/j.engappai.2025.112096}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112096}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep reinforcement learning optimization algorithm based on heterogeneous graph neural network for hybrid flow shop scheduling problem with finite transportation resources}, volume = {161}, year = {2025}, }
• (2025). Unknown fault incremental learning based on shapelet prototypical network for streaming industrial signals. EAAI, 161, 112094. (www)

  @article{ , abstract = {Unknown faults represent faults that have never occurred before, they are constantly emerging due to the changing environments and operations in industrial processes. It is a challenge for existing fault diagnosis methods to continually detect unknown faults and effectively classify known faults in streaming industrial signals. This article proposes an unknown fault incremental learning method for streaming industrial signals. In this work, shapelet prototypical embedding combined with a memory distance matrix is employed to embed streaming industrial signals into a discriminative feature space. Therefore, the category information in the signals can be extracted and is not limited by the size of the sliding window. Besides, a new training paradigm based on meta-learning by sampling simulated-incremental tasks is proposed to obtain generalizable shapelets. Moreover, based on the new training paradigm, the meta-discovery module is proposed to continually detect unknown faults, and the meta-calibrate module can calibrate all prototypes into a distinguishable space. Experiments on the simulated streaming time series, benchmark Tennessee Eastman process, and real-world aluminum electrolysis process illustrate the superiority of the proposed method in terms of accuracy and interpretability. The code is available in https://github.com/XiaoxueWan/UFIL.git .}, archive = {J_EAAI}, author = {Xiaoxue Wan and Lihui Cen and Xiaofang Chen and Yongfang Xie and Zhaohui Zeng}, doi = {10.1016/j.engappai.2025.112094}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112094}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unknown fault incremental learning based on shapelet prototypical network for streaming industrial signals}, volume = {161}, year = {2025}, }
• (2025). Adaptive attention graph convolution network with normalized embedded gaussian for rapid serial visualization presentation decoding. EAAI, 161, 112093. (www)

  @article{ , abstract = {Graph convolutional networks (GCNs) have been widely used in Brain Computer Interface (BCI) and have shown great prowess in identifying electroencephalogram (EEG) spatiotemporal features. However, most GCNs learn channels topological relationship by fixed adjacency matrix. This lacks connectivity strength information and ignores the data dependency. This paper proposes a data-driven adjacency matrix based on normalized embedded Gaussian function, and constructs a Gaussian-Adaptive Attention Graph Convolution Network (Gaussian-AAGCN). Brain regions connectivity is calculated by normalized embedded Gaussian function, and the topological relationship is adaptively learned by input data in a data-driven manner. This data-driven adaptive adjacency matrix avoids brain activity information loss caused by fixed adjacency matrix and improves the flexibility of graph construction. Convolutional block attention module (CBAM) is introduced to adaptive feature refinement in two independent dimensions, improving model representation ability. Experimental results show that the average area under curve (AUC), true positive rate (TPR) and false positive rate (FPR) of Gaussian-AAGCN on 14 subjects are 93.52 %, 91.59 %, and 4.58 % respectively. Compared to Transformer, Event-Related Potential Capsule Network (ERP-CapsNet), Electroencephalogram Convolutional Neural Network (EEGNet), and Multi-Granularity Information Fusion Network (MGIFNet), the AUC of Gaussian-AAGCN is higher by 18.02 %, 5.32 %, 2.62 %, and 1.82 %, respectively. Using the adaptive adjacency matrix, the model AUC and TPR are increased by about 4.8 % and 7.8 % respectively. After integrating CBAM, the AUC and TPR increased by about 3.5 % and 8 % respectively.}, archive = {J_EAAI}, author = {Mengyuan Zhao and Qingsong Ai and Kun Chen and Quan Liu and Sheng Quan Xie and Li Ma}, doi = {10.1016/j.engappai.2025.112093}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112093}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive attention graph convolution network with normalized embedded gaussian for rapid serial visualization presentation decoding}, volume = {161}, year = {2025}, }
• (2025). GREEN: Graph reasoning enhanced encoder network for social intention-aware forecast of vessel navigating trajectory. EAAI, 161, 112092. (www)

  @article{ , abstract = {Recent studies have demonstrated that graph neural network (GNN) excel in interactive modeling, particularly by enhancing reasoning capabilities through the construction of interaction graphs, and the application of graph convolutional network (GCN). However, these methods remain heavily dependent on prior features, which limits the ability of vessels to accurately infer each other’s decision-making intentions in encounter situations. Therefore, this paper proposes graph reasoning enhanced encoder network (GREEN), which enables social community insight on multi-ships interactions, and latent intention aware for interactive trajectory forecast. The architecture of GREEN includes two layers of trajectory-graph-embedded encoders layer (TGEEL) and multi-graphs-gated fusion layer (MGGFL). In TGEEL, we respectively design intention trajectory generator unit for graph-embedded multi-ships trajectory representation via random walk process, and latent intention generator unit to forecast future navigating intentions via variational autoencoder. In MGGFL, we also design two units, where one unit is in charge of social aware construction and spatio-temporal features extraction based on interactive multi-graphs, and the other one is responsible for gated fusion of multi-scale features and forecast future trajectory using spatio-temporal attention. Experimental results across five datasets were collected from real-world maritime environments, which demonstrated that GREEN achieved improvements of 39.98%, 38.57%, 38.65%, and 63.33% in average displacement error , final displacement error, maximum displacement error, and miss rate, compared with state-of-the-art methods. The paper highlights GREEN robust potential in complex navigational scenarios, and provides pivotal support for advancing the development of intelligent maritime systems.}, archive = {J_EAAI}, author = {Junhao Jiang and Yi Zuo}, doi = {10.1016/j.engappai.2025.112092}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112092}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GREEN: Graph reasoning enhanced encoder network for social intention-aware forecast of vessel navigating trajectory}, volume = {161}, year = {2025}, }
• (2025). DiffAT: Effective data augmentation with diffusion models for time series forecasting. EAAI, 161, 112091. (www)

  @article{ , abstract = {Data augmentation offers a promising solution to data scarcity in deep learning-based time series forecasting. However, current approaches face dual limitations (1) Hand-designed methods (e.g., cropping/masking): often disrupt the continuity of vital temporal patterns (such as seasonal and trend) by introducing abrupt pattern discontinuities; (2) Generative models often face difficulties in preserving task-critical features that are essential for prediction, especially when aiming to generate diverse augmented series. To tackle these dilemmas, we propose a novel conditional diffusion-based data augmentation framework, named DiffAT, for time series forecasting tasks. DiffAT synergizes: (1) Patch-wise masking reconstruction to capture structural invariants (such as autocorrelation and causality), and (2) encoding hand-designed augmentation prototypes for guiding diversity-preserving generation. DiffAT achieves dual enhancement: maintaining continuity of temporal patterns through progressive denoising process and exposing latent invariant patterns via guided diversity injection. We validate the efficacy of DiffAT through extensive experiments on seven real-world datasets, by comparing DiffAT with six state-of-the-art time series data augmentation methods. The results indicate our method can boost the forecasting performance of Autoformer by up to 6.49 % in 26/28 cases and improve forecasting performance of LightTS by up to 3.11 % in 23/28 cases on 7 real-world benchmarks. Extensive experiments also indicate that DiffAT can improve the accuracy of forecasting models in few-shot scenario (with 1 % training data) in 54/60 cases. We will release the source code upon publication.}, archive = {J_EAAI}, author = {Yang Yu and Ruizhe Ma and Wenbo Gu and Zongmin Ma}, doi = {10.1016/j.engappai.2025.112091}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112091}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DiffAT: Effective data augmentation with diffusion models for time series forecasting}, volume = {161}, year = {2025}, }
• (2025). Evaluating reinforcement learning-based neural controllers for quadcopter navigation in windy conditions. EAAI, 161, 112090. (www)

  @article{ , abstract = {Accurate quadcopter navigation under windy conditions remains challenging for traditional control methods, especially in the presence of unpredictable wind gusts and strict navigational constraints. This paper evaluates Deep Reinforcement Learning (DRL) based controllers under such conditions, analysing the impact of wind domain randomisation, multi-goal training, enhanced state representations with explicit wind information, and the use of temporal data to capture affecting dynamics over time. Experiments in the AirSim simulator across four trajectories — evaluated under both no-wind and windy conditions — demonstrate that DRL-based controllers outperform classical methods, particularly under stochastic wind disturbances. Moreover, we show that training a DRL agent with domain randomisation improves robustness against wind but reduces efficiency in no-wind scenarios. However, incorporating wind information into the agent’s state space enhances robustness without sacrificing performance in wind-free settings. Furthermore, training with stricter waypoint constraints emerges as the most effective strategy, leading to precise trajectories and improved generalisation to wind disturbances. To further interpret the learned policies, we apply Shapley Additive explanations analysis, revealing how different training configurations influence the agent’s feature importance. These findings underscore the potential of DRL-based neural controllers for resilient autonomous aerial systems, highlighting the importance of structured training strategies, informed state representations, and explainability for real-world deployment.}, archive = {J_EAAI}, author = {Alain Andres and Aritz D. Martinez and Sümer Tunçay and Ignacio Carlucho}, doi = {10.1016/j.engappai.2025.112090}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112090}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluating reinforcement learning-based neural controllers for quadcopter navigation in windy conditions}, volume = {161}, year = {2025}, }
• (2025). Pixel-level semantics boosted fine-grained bird image classification. EAAI, 161, 112089. (www)

  @article{ , abstract = {Fine-grained bird image classification (FBIC) is crucial for endangered bird conservation and biodiversity research. However, existing methods often struggle to capture detailed features and manage the interference caused by complex backgrounds. To address these challenges, we propose a novel Pixel-Level Semantic Boosted Fine-Grained Bird Image Classification (PFIC) framework, which enhances fine-grained bird image classification by incorporating pixel-level semantic information. PFIC consists of two core components: the Grouped Detail Enhancement (GDE) module and the Background–Foreground Enhancement (BFE) strategy. GDE integrates multi-level pixel-level semantic information, derived from a segmentation feature extractor, into classification features via two submodules: grouped aggregation and detail enhancement. This approach enhances the model’s ability to capture fine-grained details. BFE augments training samples by restricting background ranges and applying random shifts to foreground objects, thereby improving the model’s capability to recognize foreground objects in complex environments. Experimental results demonstrate that our method achieves state-of-the-art performance on the CUB-200-2011 and NABirds datasets. Additionally, further experiments on the Stanford Cars dataset validate the framework’s potential for generalization to other fine-grained image classification tasks.}, archive = {J_EAAI}, author = {Haoxiang Ma and Yongjian Deng and Bochen Xie and Jian Liu and Hai Liu and Youfu Li and Zhen Yang}, doi = {10.1016/j.engappai.2025.112089}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112089}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pixel-level semantics boosted fine-grained bird image classification}, volume = {161}, year = {2025}, }
• (2025). Transformer-based soft sensor of five day biochemical oxygen demand indicators in the municipal wastewater treatment process. EAAI, 161, 112088. (www)

  @article{ , abstract = {Five day biochemical oxygen demand (BOD 5 ) is usually used to measure whether wastewater treatment meets the standard. Previously, the development of machine learning technology was applied to predict BOD 5 for future planning of wastewater treatment. In real wastewater treatment, the monitoring of BOD 5 often lags behind for a long time, so neural network-based soft sensor methods are widely used in this field. However, the currently used BOD 5 soft sensor model has the problem of ignoring the synergistic effect of the model and stable information transmission. In order to monitor the BOD 5 concentration more accurately, a Transformer network model with embedded Long Short-Term Memory (LSTM) network is proposed in this article : LSTM is embedded into the calculation of the transformer multi-head attention, and the output method of the multi-head attention is improved so that the attention heads are output one by one in a serial manner, so it is named LSTM-Serial-Transformer (LSformer). Compared with previous models, LSformer optimizes input information through LSTM and Transformer to achieve synergistic effect, thus better capturing the characteristics of BOD 5 time series data, while reducing the risk of gradient vanishing or gradient exploding during data transmission to improve the stability of the model. Finally, the model was verified to have better soft sensor performance on a benchmark simulation model and a real-world dataset.}, archive = {J_EAAI}, author = {Xiaoling Zhang and Zhi Qi and Peng Chang and Zhiqi Hu}, doi = {10.1016/j.engappai.2025.112088}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112088}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based soft sensor of five day biochemical oxygen demand indicators in the municipal wastewater treatment process}, volume = {161}, year = {2025}, }
• (2025). ForgetMe: Benchmarking the selective forgetting capabilities of generative models. EAAI, 161, 112087. (www)

  @article{ , abstract = {The widespread adoption of diffusion models in image generation has increased the demand for privacy-compliant unlearning. However, due to the high-dimensional nature and complex feature representations of diffusion models, achieving selective unlearning remains challenging, as existing methods struggle to remove sensitive information while preserving the consistency of non-sensitive regions. To address this, we propose an Automatic Dataset Creation Framework based on prompt-based layered editing and training-free local feature removal, constructing the ForgetMe dataset and introducing the Entangled evaluation metric. The Entangled metric quantifies unlearning effectiveness by assessing the similarity and consistency between the target and background regions and supports both paired ( Entangled-D ) and unpaired ( Entangled-S ) image data, enabling unsupervised evaluation. The ForgetMe dataset encompasses a diverse set of real and synthetic scenarios, including CUB-200-2011 (Birds), Stanford-Dogs, ImageNet, and a synthetic cat dataset. We apply LoRA fine-tuning on Stable Diffusion to achieve selective unlearning on this dataset and validate the effectiveness of both the ForgetMe dataset and the Entangled metric, establishing them as benchmarks for selective unlearning. Our work provides a scalable and adaptable solution for advancing privacy-preserving generative AI. Code is available at: https://github.com/YuZhenyuLindy/ForgetMe .}, archive = {J_EAAI}, author = {Zhenyu Yu and Mohd Yamani Idna Idris and Pei Wang and Yuelong Xia and Yong Xiang}, doi = {10.1016/j.engappai.2025.112087}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112087}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ForgetMe: Benchmarking the selective forgetting capabilities of generative models}, volume = {161}, year = {2025}, }
• (2025). A novel data generation scheme for surrogate modelling with deep operator networks. EAAI, 161, 112086. (www)

  @article{ , abstract = {Operator-based neural network architectures such as deep operator networks have emerged as a promising tool for the surrogate modelling of physical systems. In general, the training data for operator surrogate modelling are generated by solving partial differential equations using the finite element method. The computationally intensive nature of data generation is one of the major bottlenecks in deploying these surrogate models, hindering the deployment of these surrogate models in practical applications. In this study, we propose a novel methodology to alleviate the computational burden associated with training data generation for deep operator networks. Unlike the existing literature, the proposed framework for data generation does not use any partial differential equation integration strategy, such as the finite element method. In the proposed strategy, the primary field consistent with the boundary conditions is first generated randomly using Gaussian process regression. Thereafter, from the primary field, the input source field is calculated using second-order accurate finite difference techniques. The computation of the derivatives from the finite difference scheme is significantly less computationally intensive and easier to implement compared to integrating the underlying governing equations, thereby reducing the computational cost associated with generating training datasets. To validate the proposed approach, we employ heat equations as a model problem and develop the surrogate model for numerous boundary value problems.}, archive = {J_EAAI}, author = {Shivam Choubey and Shailendra Rahi and Birupaksha Pal and Manish Agrawal}, doi = {10.1016/j.engappai.2025.112086}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112086}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel data generation scheme for surrogate modelling with deep operator networks}, volume = {161}, year = {2025}, }
• (2025). Interpretable generalization diagnosis: Variational causal disentanglement-based coalitional game attribution network. EAAI, 161, 112084. (www)

  @article{ , abstract = {Interpretability and controllability are the keys to intelligent fault diagnosis in industrial scenarios. However, most intelligent methods have focused on post-hoc feature importance explanations, which cannot constrain the uncontrollable learning behavior or form an interpretable diagnosis process consistent to match expert experience. To address this problem, this paper proposes a variational causal disentanglement-based coalitional game attribution network (VCN) to achieve a controllable and interpretable modeling paradigm that can draw diagnostic conclusions with conforming to expert experience. This model enables independent causal factor disentanglement and model regularization during the training process, as well as fast explanation computation at test time. Variational causal disentanglement is defined and do-operations and variational inference processes are introduced to generate training data with independent causal factors for subsequent attribution. The coalitional game attribution establishes a coalitional game relationship between the model and the independent causal factors, and regularizes the model so that it converges to a state that conforms to logical axioms. Both laboratory and real-world data demonstrate the proposed method enables the provision of interpretable diagnostic conclusions while maintaining controllable performance. The code is available at: https://github.com/cyber-dogy/VCN-code .}, archive = {J_EAAI}, author = {Junwei Gu and Changhua Hu and Yu Wang and Mingquan Zhang and Yanzhuo Lin and Shangjing Peng and Dongdong Li}, doi = {10.1016/j.engappai.2025.112084}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112084}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable generalization diagnosis: Variational causal disentanglement-based coalitional game attribution network}, volume = {161}, year = {2025}, }
• (2025). A novel photovoltaic cell defect detection method: A deep learning model based on multi-scale enhanced feature extraction. EAAI, 161, 112083. (www)

  @article{ , abstract = {With the increasingly widespread application of clean energy technology, the global demand for photovoltaic cells is gradually increasing, and the status and defects of photovoltaic cells are also being taken seriously. However, there are various types of defects in photovoltaic cells, including those that are difficult to detect, making detection work somewhat challenging. This article proposes a novel improved model for defect detection in photovoltaic cells based on You Only Look Once (YOLO). Firstly, the Multi-Head Self-Attention (MHSA) mechanism is adopted to compensate for the shortcomings of the Cross Stage Partial Bottleneck with 2 Convolutions (C2f) module in channel feature extraction. Secondly, the Receptive-Field Coordinate Attention Convolutional (RFCAConv) operation module is utilized to expand the receptive field of defect detection in the model and increase the range of feature extraction. Finally, the Bi-directional Feature Pyramid Network (BiFPN) and Concat modules are integrated to enhance the feature relationships between the output connections of each module and strengthen the output results. The final experimental results demonstrate that the improved model achieves an accuracy of 90.3%, representing a 5.2% increase compared to the original model. Through experimental verification, the improved model proposed in this paper can not only detect defects in photovoltaic cells in electroluminescence (EL) images but also in infrared thermal images. The model has strong generalization ability.}, archive = {J_EAAI}, author = {Yu Gao and Zhanying Li and Yinghao Zhang and Kangye Zhang and Haoyang Yu}, doi = {10.1016/j.engappai.2025.112083}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112083}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel photovoltaic cell defect detection method: A deep learning model based on multi-scale enhanced feature extraction}, volume = {161}, year = {2025}, }
• (2025). Wind power prediction based on hybrid deep learning and monte carlo simulation. EAAI, 161, 112082. (www)

  @article{ , abstract = {This study proposes a hybrid deep learning model combining Variational Mode Decomposition (VMD), Convolutional Neural Networks (CNN), and a Three-Dimensional Gated Neural Network (TGNN) to enhance wind power prediction accuracy. VMD decomposes wind power time series into intrinsic mode functions, CNN extracts deep features, and TGNN captures temporal dependencies with error feedback control. A Monte Carlo simulation based on the Central Limit Theorem is introduced to evaluate predictive uncertainty and interval coverage. Experimental results from three wind farms in China demonstrate that the proposed model significantly outperforms baseline hybrid models. Specifically, it achieves a coverage probability (CP) of 0.887, an average interval width (AIW) of 200.870, and a normalized mean square error (NMSE) of 0.057. Compared with the VMD-CNN-LSTM model, the proposed VMD-CNN-TGNN reduces root mean error (RMSE) by approximately 50%, indicating its superior accuracy, robustness, and practical value in wind power forecasting.}, archive = {J_EAAI}, author = {Zhiyong Guo and Qiaoli Han and Fangzheng Wei and Wenkai Qi}, doi = {10.1016/j.engappai.2025.112082}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112082}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Wind power prediction based on hybrid deep learning and monte carlo simulation}, volume = {161}, year = {2025}, }
• (2025). A parallel memetic algorithm for qubit mapping on noisy intermediate-scale quantum machines. EAAI, 161, 112081. (www)

  @article{ , abstract = {Quantum computer technologies are advancing rapidly, but current devices are still limited to the realm of Noisy Intermediate Scale Quantum (NISQ) systems. Due to the limited set of gates and scarce physical qubit connectivity, most quantum circuit-based programs need to be transpiled to be executed on such hardware. This transpilation, which consists in converting the quantum circuit into a hardware-compliant one includes the critical qubit mapping step, where logical qubits are mapped onto physical qubits to meet the hardware connectivity constraints. In this paper, we investigate the application of Artificial Intelligence (AI) to the qubit mapping problem. More exactly, we propose a Parallel Memetic Algorithm for Qubit Mapping (PMA-QM), designed to tackle this challenge efficiently. Using a fine-tuned parallel model to accelerate this inherently computationally expensive hybrid approach, our algorithm takes problem-specific knowledge into account to optimize circuit depth, reducing execution time, and minimizing error rates consequently. PMA-QM has been experimented using various medium-to-large scale circuit benchmarks. PMA-QM outperforms the widely used SWAP-based BidiREctional (SABRE) algorithm, delivering consistently better solutions.}, archive = {J_EAAI}, author = {Jérôme Rouzé and Nouredine Melab and Jan Gmys and Daniel Tuyttens}, doi = {10.1016/j.engappai.2025.112081}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112081}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A parallel memetic algorithm for qubit mapping on noisy intermediate-scale quantum machines}, volume = {161}, year = {2025}, }
• (2025). Enhancing non-orthogonal multiple access systems: A reconfigurable machine learning classification approach. EAAI, 161, 112080. (www)

  @article{ , abstract = {Non-Orthogonal Multiple Access (NOMA) is being considered as a Multiple Access (MA) technique for the next-generation systems, due to factors such as a high per-user spectral resource allocation, grant-free transmission, support for millimeter Waves (mmWaves) and massive-MIMO (mMIMO). A practical limitation of the NOMA systems is imperfect Successive Interference Cancellation (SIC); and the involved decoding delay. This work develops a data-driven Machine Learning (ML) model providing the functionality of a SIC receiver. While most of the Deep Learning (DL) algorithms have high complexity and training times, the given ML approach utilizes the received symbol as the only primary predictor. The other two predictors are derived from the primary predictor, and utilized for the Power-Domain (PD)-NOMA symbol-decoding process. The model utilizes a developed low-complexity NOMA-ML-based Decoder (MLbD) dataset for the same. The extensive test simulations confirm the reconfigurable ML-based receiver to be at par with the existing Maximum Likelihood (MLH) decoder in terms of decoding accuracy. Still, the former supports its integration into the next-generation systems due to its reconfigurable nature and removes the drawbacks related to SIC in the NOMA systems.}, archive = {J_EAAI}, author = {Saurabh Srivastava and Rampravesh Kumar and Prajna Parimita Dash}, doi = {10.1016/j.engappai.2025.112080}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112080}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing non-orthogonal multiple access systems: A reconfigurable machine learning classification approach}, volume = {161}, year = {2025}, }
• (2025). Collaborative energy-saving path planning of unmanned surface vehicle cluster based on multi-head attention mechanism and multi-agent deep reinforcement learning. EAAI, 161, 112078. (www)

  @article{ , abstract = {This paper proposes a Multi-Agent Deep Reinforcement Learning (MADRL) algorithm to address the overlooked issue of energy consumption optimization in existing cluster path planning methods, enabling collaborative energy-efficient path planning for Unmanned Surface Vehicle (USV) clusters. Meanwhile, it provides an alternative MADRL approach to address practical engineering challenges. We not only consider the problems of avoiding danger, reaching the target point and avoiding path conflicts between USV, but also further consider the energy consumption and speed planning of USV cluster. Specifically, firstly, we establish the USV motion model and the USV cluster conflict model, and propose a new energy consumption model, which considers the relationship among speed, marine environment and propulsion load. Secondly, we propose a multi-head local state relevance capture mechanism-multi-agent proximal policy optimization (MLSRC-MAPPO) algorithm. This algorithm can use multi-head attention mechanism (MHA) to capture the potential dependencies between USV, thus enhancing the convergence performance of multi-agent. Finally, in order to reduce the training difficulty, we propose a multi-dimensional action space method for action networks. The experimental results demonstrate that the proposed multi-dimensional action space method has achieved significant success in the lightweighting of the action network: the number of network parameters was reduced by 89.10%, and the computational complexity was decreased by 88.94%, significantly enhancing training efficiency. Meanwhile, the MLSRC-MAPPO algorithm, by incorporating a multi-head attention mechanism, greatly improved the convergence performance of the multi-agent system. In test scenarios with both identical and different starting points, the method reduced energy consumption by 26.24% and 38.47% respectively, fully validating its effectiveness and superiority. Furthermore, comparative experiments with existing cluster energy-saving path planning methods show that the proposed method exhibits clear advantages in terms of energy consumption and path planning efficiency, further verifying its superiority. The corresponding code for this paper is as follows: https://github.com/xhpxiaohaipeng/Multi_USV_Trajectory_energy_plan}, archive = {J_EAAI}, author = {Haipeng Xiao and Lijun Fu and Chengya Shang and Yunfeng Lin and Longfei Yue and Yaxiang Fan}, doi = {10.1016/j.engappai.2025.112078}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112078}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative energy-saving path planning of unmanned surface vehicle cluster based on multi-head attention mechanism and multi-agent deep reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Identification of surface subsidence risk in deep foundation pits using a mamba fusion model. EAAI, 161, 112077. (www)

  @article{ , abstract = {This study proposes a novel artificial intelligence–driven neural network model, CNN-Mamba-LSTM-SA, for predicting surface settlement induced by deep foundation pit excavation. The model integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a Self-Attention (SA) mechanism, and the Mamba architecture to capture both spatial and long-range temporal dependencies in multi-source monitoring data. Bayesian optimization is employed for hyperparameter tuning, and Variational Mode Decomposition (VMD) is used for data denoising, resulting in improved prediction accuracy. To enhance model interpretability, Shapley Additive Explanations (SHAP) are applied to identify key deformation drivers, revealing groundwater level and building settlement as the most influential factors. Model performance is validated using monitoring data from the Nanjing Gemini excavation project, where it achieves superior results compared to conventional models. The CNN-Mamba-LSTM-SA model reduces Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) by up to 69.77 %, 62.45 %, and 79.45 %, respectively. Further analysis through ablation experiments confirms the contribution of each module. Interestingly, the combined removal of CNN and SA results in greater performance degradation than the sum of their individual effects. Finally, a risk warning framework is developed to enable the dynamic transformation of predicted and observed settlement values into actionable risk levels. This integrated artificial intelligence approach offers a robust and interpretable tool for managing excavation-induced risks in urban geotechnical engineering.}, archive = {J_EAAI}, author = {Chenhe Ge and Pengfei Li and Mingju Zhang and Meng Yang}, doi = {10.1016/j.engappai.2025.112077}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112077}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification of surface subsidence risk in deep foundation pits using a mamba fusion model}, volume = {161}, year = {2025}, }
• (2025). Risk management in maintenance processes: A spherical fuzzy-based failure mode and effect analysis approach in the glass processing industry. EAAI, 161, 112076. (www)

  @article{ , abstract = {Quality management systems are essential tools that aim to increase the competitiveness of businesses and ensure customer satisfaction by creating reliable quality control processes. In this study, the importance of the "risk-based thinking" approach of the International Organization for Standardization (ISO) 9001:2015 standard within the scope of quality management systems was emphasized, and a risk assessment was made for the maintenance process of a business in the glass processing sector. In order to eliminate the deficiencies of the failure mode and effect analysis (FMEA) method, the integration of spherical fuzzy sets (SFSs) and the Technique of Order Preference Similarity to the Ideal Solution (TOPSIS) method was used. Within the scope of the study, 17 different risks were determined by five experts from the maintenance, repair, and quality assurance departments. The experts evaluated these identified risks according to occurrence, severity, and detection factors. Then, these risks were ranked using the Spherical Fuzzy TOPSIS (SF-TOPSIS) method. Finally, different scenarios were created, and their results were discussed to provide a more comprehensive and sensitive risk management approach. This study focuses on maintenance processes in the glass processing industry as a case study. However, the risks related to the maintenance process defined in the study (e.g., machine failures, maintenance inefficiencies, spare parts shortages) are common in the manufacturing sector. Therefore, it can be applied to the sector where the application was carried out and to many different sectors and enterprises.This methodology also serves as a guide for businesses that want to manage process risks within the scope of ISO 9001:2015.}, archive = {J_EAAI}, author = {Buse Duygu Dağıdır and Barış Özkan}, doi = {10.1016/j.engappai.2025.112076}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112076}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk management in maintenance processes: A spherical fuzzy-based failure mode and effect analysis approach in the glass processing industry}, volume = {161}, year = {2025}, }
• (2025). Few-shot augmentation based on variational auto-generative adversarial network with moving losses: Application to the variable stiffness prediction in composites. EAAI, 161, 112075. (www)

  @article{ , abstract = {The dataset of carbon fiber reinforced plastics (CFRP) materials presents characteristics of high-dimensional, low-rank, and sparse, which pose difficulties in the combination of mechanical modeling. In this paper, a Variational Auto-Generative Adversarial Network (VAGAN) with moving losses is proposed as a data augmentation method, which extends the size of the CFRP dataset covering components, processes, elasticity, and strengths factors, and increases the information conveyed in the surrogate modeling. A compression strength prediction model for CFRP laminates was constructed by combining the component, process, and mechanical tensor with a neural network optimized by the search algorithm. Combined with the data augmentation strategy, not only was the amount of data expanded, but the prediction accuracy was also significantly improved. The allowable value of compression strength is analyzed and calculated by the predicted values, which brings direct benefits in simplifying the test.}, archive = {J_EAAI}, author = {Zhicen Song and Yunwen Feng and Cheng Lu and Jiaqi Liu}, doi = {10.1016/j.engappai.2025.112075}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112075}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Few-shot augmentation based on variational auto-generative adversarial network with moving losses: Application to the variable stiffness prediction in composites}, volume = {161}, year = {2025}, }
• (2025). Multivariate time series forecasting based on temporal decomposition and graph neural network. EAAI, 161, 112074. (www)

  @article{ , abstract = {It is quite challenging to forecast the Multivariate Time Series (MTS) accurately due to the high dimensionality of MTS and the entangled correlation between variables. Recently, graph-based networks have been demonstrated to be an effective model to handle the complex correlations between MTS. However, all existing graph-based methods construct the graph model of the MTS using only the shallow correlations from the raw MTS data, ignoring the deep-rooted correlations hidden in the features. In this paper, we propose for the first time to construct a comprehensive graph model of MTS that incorporates both shallow correlations from raw data and hidden correlations from decomposed temporal properties. Then, we propose a novel graph-based MTS forecasting framework, which optimizes the graph structure jointly with the model parameters. By doing so, the graph structure can adaptively model the correlations of MTS at a deep level, while the joint optimization can make the constructed graph compatible with the forecasting tasks of MTS, contributing to a globally optimal solution. Finally, we conduct extensive experiments on seven real-world datasets, the results demonstrate the superiority of our method on MTS forecasting over the state-of-the-art baselines. The source codes of the experiments with datasets are available at https://github.com/ironweng/MF-TDGNN .}, archive = {J_EAAI}, author = {Yan Qiao and Pei Zhao and Junjie Wang and Rongyao Hu and Minyue Li and Xinyu Yuan and Meng Li and Zhenchun Wei and Cuiying Feng}, doi = {10.1016/j.engappai.2025.112074}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112074}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multivariate time series forecasting based on temporal decomposition and graph neural network}, volume = {161}, year = {2025}, }
• (2025). A unified framework for lung disease screening using chest X-ray images. EAAI, 161, 112073. (www)

  @article{ , abstract = {Chest X-ray (CXR) imaging is a crucial diagnostic tool for identifying respiratory diseases. The existing methods face challenges such as misclassification caused by overlapping radiologic patterns and the shortage of trained radiologists. However, previous studies have utilized deep learning (DL) models to overcome these limitations, but they struggle with imbalanced datasets, and noise sensitivity. In addition, most of the studies are based on the binary classification of CXR images. Thus, the proposed framework integrates segmentation and multi-class classification for improved generalization and robustness. In this work, artificial intelligence techniques like encoder–decoder and ensemble learning are utilized for segmentation and classification of CXR images that offers reliable, and efficient solution. The output of the presented framework classifies the CXR images into four different classes, along with their corresponding class score. The proposed classification model is ensemble of three pre-trained DL architectures. The last five layers of these base models are fine tuned to match the classification process and combined through a multi-layer perceptron classifier for improved accuracy. The proposed model achieves overall accuracy of 88.98%, and area under curve value of 0.9753, outperforming several state-of-the-art models. It is a lightweight model and demonstrates significant robustness under noisy conditions compared to other models. Moreover, the proposed segmentation and ensemble models are trained and tested on different datasets to obtain greater robustness. The data augmentation is also employed to address class imbalance nature of dataset and enhance the generalization capability. Further, statistical analysis is performed to present the comprehensive comparison among the models.}, archive = {J_EAAI}, author = {Bhavana Singh and Pushpendra Kumar}, doi = {10.1016/j.engappai.2025.112073}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112073}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified framework for lung disease screening using chest X-ray images}, volume = {161}, year = {2025}, }
• (2025). A hybrid bayesian model updating and non-dominated sorting genetic algorithm framework for intelligent mix design of steel fiber reinforced concrete. EAAI, 161, 112071. (www)

  @article{ , abstract = {Steel fiber reinforced concrete (SFRC) improves the strength and toughness of conventional concrete, but the high cost and carbon footprint of fibers challenge the balance among performance, cost and sustainability. To address this, an intelligent mix design framework is proposed to optimize compressive and splitting tensile strengths, cost and emissions. Based on 671 experimental records, posterior models were built using Markov Chain Monte Carlo sampling and Bayesian model updating, enabling accurate strength predictions. Compared to traditional regression methods, R 2 scores improved by 15.7 % and 12.4 %, confirming its predictive advantage. Cost-wise, materials dominate, while emissions mainly arise from production, transport and mixing. A non-dominated sorting genetic algorithm identified optimal designs under given constraints. Results show that reducing water-to-cement and aggregate-to-cement ratios, and increasing sand ratio and fiber reinforcement index, enhances SFRC strength. Larger coarse aggregates reduce compressive strength but have limited effect on tensile strength. Optimization suggests potential cost and emission reductions of up to 60 %. Moreover, for compression-prone components, fiber use is inefficient due to high cost and emissions, whereas for crack-resistant or strength-balanced elements, fiber inclusion offers a more sustainable alternative to merely lowering the water-to-cement ratio. The proposed framework enables tailored SFRC mix designs, guiding the efficient use of steel fibers.}, archive = {J_EAAI}, author = {Yong Yu and Jie Su and Bo Wu}, doi = {10.1016/j.engappai.2025.112071}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112071}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid bayesian model updating and non-dominated sorting genetic algorithm framework for intelligent mix design of steel fiber reinforced concrete}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional human pose estimation based on multi-scale spatial–temporal transformer. EAAI, 161, 112068. (www)

  @article{ , abstract = {Recently, transformer-based methods have become dominant in the domain of three-dimensional (3D) human pose estimation, yet the U-net model based on convolutional neural networks (CNN) struggles to model long temporal sequences, and sequence-to-frame (Seq2frame) and sequence-to-sequence (Seq2seq) approaches often fail to preserve dependencies at the start and end of sequences. To address these challenges, this paper proposes a Multi-Scale Spatial–Temporal Transformer network (MSST). This network utilizes Sequence Padding Module (SPM) to extract edge features of the first and last frames, and employs Spatial–Temporal Transformer (STT) to model the spatial–temporal correlations of keypoints. Additionally, we design a Multi-Scale Module (MSM) that analyzes the human skeletal topology to extract multi-scale features of keypoints, local information, and global information, and fuse semantic information at different scales. Finally, we utilize regression heads to project the processed keypoint feature information into 3D space. We conduct quantitative evaluations on two benchmark datasets using four evaluation metrics and design multiple sets of comparative experiments to validate the effectiveness of the proposed modules. Experimental results demonstrate that the proposed network achieves excellent performance.}, archive = {J_EAAI}, author = {Xiaogang Song and Yongxin Cui and Jichen Chen and Xinhong Hei}, doi = {10.1016/j.engappai.2025.112068}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112068}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional human pose estimation based on multi-scale spatial–temporal transformer}, volume = {161}, year = {2025}, }
• (2025). SharpEdge: High-quality data-driven monocular depth estimation for enhanced boundary precision. EAAI, 161, 112067. (www)

  @article{ , abstract = {While existing monocular depth estimation methods have achieved commendable performance, they often fall short in accurately distinguishing object boundaries. This deficiency largely stems from the inherent noise in dataset acquisition, such as unclear edges and missing depth information. To address these challenges, this paper introduces a novel, high-quality, data-driven monocular depth estimation method tailored for autonomous driving. The approach significantly enhances depth predictions with clearer object boundaries and reduced noise, making it well-suited for real-time, safety-critical applications. Central to our approach is the Self-Adaptive Consistency Filtering mechanism, which dynamically selects high-quality training samples, ensuring that the model learns from the most reliable data and reducing the impact of noise. Additionally, we introduce a Dual-Prior Learning strategy that combines geometric and semantic edge priors. Unlike traditional methods that rely solely on raw depth maps, our approach enhances boundary detection by providing detailed guidance on object contours. This leads to more accurate depth estimation, especially in complex regions where other methods struggle. Empirical evaluations on popular benchmark datasets show that our approach leads to performance improvements of 1.2% on the autonomous driving dataset KITTI and 1.8% on the indoor scene dataset NYU. Compared with recent state-of-the-art methods such as DPT and NewCRFs, our approach achieves superior performance, particularly in recovering fine object boundaries and maintaining spatial consistency across diverse scenes. These results highlight the strong generalization ability of our method, demonstrating that it can enhance depth estimation quality across diverse environments — improving edge precision and spatial coherence, which are critical for autonomous vehicles navigating both complex and dynamic scenarios.}, archive = {J_EAAI}, author = {Mengke Song and Luming Li and Xu Yu and Chenglizhao Chen and Shanchen Pang}, doi = {10.1016/j.engappai.2025.112067}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112067}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SharpEdge: High-quality data-driven monocular depth estimation for enhanced boundary precision}, volume = {161}, year = {2025}, }
• (2025). A systematic review on vision-based gaze estimation: Advance in computer vision and deep learning. EAAI, 161, 112066. (www)

  @article{ , abstract = {Eye gaze estimation is all about figuring out where a person is looking by analyzing eye movements. This is usually done using cameras or eye-tracking devices, a key technique in computer vision. Gaze estimation has a wide range of uses—from helping people interact with computers more naturally (HCI), monitoring drivers, enhancing virtual reality experiences, and reading emotional cues in affective computing. Deep learning has improved how accurately and reliably these systems work in recent years. Models like Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs) have made it easier to detect and interpret eye movements in complex settings. This review closely examines how gaze estimation works, the datasets commonly used, and both traditional and deep learning-based methods. It also points out what's working well, where current methods fall short, and what researchers still need to solve. The goal is to offer helpful insights for anyone working to improve or apply gaze tracking technologies in real-world scenarios.}, archive = {J_EAAI}, author = {Sapna Singh Kshatri and Deepak Singh}, doi = {10.1016/j.engappai.2025.112066}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112066}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A systematic review on vision-based gaze estimation: Advance in computer vision and deep learning}, volume = {161}, year = {2025}, }
• (2025). A novel framework for improving railway driver performance based on emotional intelligence and job-driven factors: An artificial neural network method. EAAI, 161, 112065. (www)

  @article{ , abstract = {Efficient transportation systems critically rely on human factors to ensure safety, reliability, and service quality. This study introduces a novel framework for evaluating and improving the performance of the Tehran Urban and Suburban Railway Company (TRS) drivers by focusing on the role of emotional intelligence (EI) and job-driven (JD) factors. The core of the proposed method is a two-stage hybrid approach: first, an Artificial Neural Network (ANN) is used to model driver performance using EI and JD factors as inputs and outputs, respectively; then, Data Envelopment Analysis (DEA) validates the results and benchmarks performance levels. Data were collected from 146 TRS drivers through a standardized questionnaire. Sensitivity analysis and statistical tests evaluated the impact of EI and JD factors on driver performance. Additionally, a SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis translated the findings into practical recommendations for system improvement. The findings reveal high job satisfaction and low job stress levels among TRS drivers, highlighting the significant contribution of EI and JD factors. By combining data-driven prediction, performance benchmarking, and strategic analysis, this study introduces a novel, human-centered framework for driver Performance Evaluation (PE)—one that fills methodological gaps in the literature and supports enhancing safety, efficiency, and passenger satisfaction in railway systems.}, archive = {J_EAAI}, author = {Narges Hajloo and Behnaz Salimi and Mahdi Hamid and Masoud Rabbani}, doi = {10.1016/j.engappai.2025.112065}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112065}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for improving railway driver performance based on emotional intelligence and job-driven factors: An artificial neural network method}, volume = {161}, year = {2025}, }
• (2025). An efficient fusion-based deep learning framework for land use and land cover image clustering. EAAI, 161, 112061. (www)

  @article{ , abstract = {Land use and land cover (LULC) analysis is vital for understanding spatial dynamics and informing environmental management, urban planning, and sustainable development. Traditional approaches, such as manual surveys and conventional image clustering methods, often face limitations in scalability and adaptability. This paper presents a novel deep learning framework that combines the Vision Transformer (ViT) and Variational Autoencoder (VAE) to extract complementary feature representations for LULC image clustering. The ViT tokenizes image patches to capture high-level semantic features, while the VAE models latent structures to integrate contextual and structural information. To further improve clustering performance, the framework incorporates Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction followed by k -means++ clustering, enabling a scalable and robust solution for diverse datasets. Experiments on multiple datasets, including the Urban Atlas LULC 2018 dataset and recent LULC maps of Japan and Vietnam, demonstrate the framework’s superior ability to capture complex LULC patterns compared to traditional methods. The datasets and source code will be made publicly available at https://github.com/ClarkDinh/LULCMiner . This framework has broad applications across geospatial and remote sensing engineering, civil and environmental engineering, agricultural planning, transportation, and urban development.}, archive = {J_EAAI}, author = {Tai Dinh and Dat Tran and Zdena Dobešová and Huynh Van Hong and Daniil Lisik and Rameesh Khan}, doi = {10.1016/j.engappai.2025.112061}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112061}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient fusion-based deep learning framework for land use and land cover image clustering}, volume = {161}, year = {2025}, }
• (2025). Optimizing multi-objectives urban siting of hydrogen refueling dispenser using fuzzy NSGA-II: A case study in fez, morocco. EAAI, 161, 112060. (www)

  @article{ , abstract = {To support the growing need for sustainable urban transport, it is essential to reduce the environmental impact of vehicles; this study focuses on optimizing hydromobility through the strategic integration of hydrogen dispensers (HDs) into existing locations, rather than deploying standalone hydrogen refueling stations (HRSs). This approach minimizes costs while enhancing accessibility, addressing the current limitations in hydrogen vehicle (HV) adoption. A novel method combining the non-dominated sorting genetic algorithm II (NSGA-II) with fuzzy logic and Pareto front analysis is proposed to identify optimal HD locations using a real case study of Fez city in Morocco. The results provide actionable strategies for implementing HDs in a way that balances efficiency, accessibility, and budget constraints. The study demonstrates how intelligent planning can support the transition to cleaner energy solutions in urban mobility.}, archive = {J_EAAI}, author = {Soukayna Abibou and Dounia El Bourakadi and Ali Yahyaouy and Hamid Gualous}, doi = {10.1016/j.engappai.2025.112060}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112060}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing multi-objectives urban siting of hydrogen refueling dispenser using fuzzy NSGA-II: A case study in fez, morocco}, volume = {161}, year = {2025}, }
• (2025). Synthetic data enhancement using diffusion models for improved miscanthus identification and bioenergy extraction. EAAI, 161, 112059. (www)

  @article{ , abstract = {Miscanthus, a high-yielding perennial grass pivotal for bioenergy production, requires precise species identification to optimize bioenergy extraction. However, limited annotated spectral datasets hinder robust classification model development. To address this challenge, the paper proposes a novel diffusion probabilistic model tailored for near-infrared spectral synthesis. Unlike conventional generative approaches, the proposed model integrates a bidirectional gated recurrent unit-based temporal encoder and one dimension convolutional neural networks within a diffusion framework, augmented by a spectral attention module to prioritize critical absorption bands. This architecture uniquely addresses the sequential dependencies and subtle biochemical variations inherent in near-infrared spectral, enabling high-fidelity generation of diverse synthetic data. The diffusion process is optimized through a hybrid loss function combining variational lower bound training with mean squared error for pixel-level fidelity and maximum mean discrepancy for distributional alignment. Evaluated on 517 near-infrared spectral samples across three Miscanthus species, the proposed model outperforms traditional variational autoencoders , generative adversarial networks, and standard diffusion models in terms of sample authenticity and diversity. Incorporating synthetic data enhanced the accuracy, precision, and recall of downstream classifiers by 10%–15%, with the convolutional neural networks attaining 87% accuracy using hybrid real-synthetic training data. Remarkably, even with 50% synthetic data substitution, classification accuracy remained robust at 75%, demonstrating the model’s efficacy in mitigating data scarcity and advancing precision agriculture for bioenergy optimization.}, archive = {J_EAAI}, author = {Xinyue Wang and Xiangdong Chen and Jun Jiang and Ronggao Gong and Biao Wang}, doi = {10.1016/j.engappai.2025.112059}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112059}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic data enhancement using diffusion models for improved miscanthus identification and bioenergy extraction}, volume = {161}, year = {2025}, }
• (2025). Multi-order feature learning with global–local attention for real-time X-ray security inspection. EAAI, 161, 112058. (www)

  @article{ , abstract = {X-ray security inspection in high-throughput environments such as airports faces fundamental challenges stemming from the intrinsic complexity of baggage scanning. These challenges include feature entanglement from overlapping objects, visual distortions due to fixed imaging angles, inconsistent pseudo-color representations caused by varying X-ray absorption, and the difficulty of detecting prohibited items across multiple scales. To tackle these challenges, we introduce Multi-Order Gated Network (MoGNet), a lightweight transformer-based architecture that integrates three core innovations. First, a Multi-order Gated Aggregation Block designed for efficient multi-order feature extraction. Second, a Global–Local Self-Attention mechanism that enhances differentiation between foreground and background elements. Finally, a DynamicFusion module for adaptive integration of multi-scale features. Comprehensive evaluations on five challenging datasets establish our proposed framework, MoGNet, as the new state-of-the-art (SOTA). The model demonstrates superior performance, achieving mean Average Precision (mAP) scores at a 50% Intersection over Union (IoU) threshold (mAP 50 ) of 75.4%, 91.9%, 91.2%, 96.6%, and 80.0%, respectively. This high accuracy is maintained while operating at an efficient 64.1 Frames Per Second (FPS). These comprehensive experimental results demonstrate its remarkable capability to optimize the balance between computational efficiency and detection accuracy, establishing it as a preferable solution for real-time contraband detection in practical security screening.}, archive = {J_EAAI}, author = {Ling Guo and Yangbin Xu and Shouhong Chen}, doi = {10.1016/j.engappai.2025.112058}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112058}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-order feature learning with global–local attention for real-time X-ray security inspection}, volume = {161}, year = {2025}, }
• (2025). A stress-dependent strength neural network model for predicting the true triaxial strength of rocks. EAAI, 161, 112057. (www)

  @article{ , abstract = {Understanding the true triaxial strength of rocks is essential for safe underground engineering, yet existing empirical and data-driven models often fail to capture the nonlinear effects of the intermediate principal stress σ 2 . This study proposes a stress-dependent strength neural network (SDSNN) that integrates physically informed constraints, including monotonicity and boundary conditions, as well as an exponential adaptive weighting strategy to balance data and constraint losses. Cohesion and internal friction angle are used as input features, replacing conventional reliance on uniaxial strength. Compared with a purely data-driven neural network (DDNN) and a constraint-aware variant without adaptive weighting (SDSNN#0), SDSNN achieves significantly better predictive accuracy and robustness across test sets, representative rock types, and five-fold cross-validation. In particular, it maintains consistent strength trends and improved stability across six representative rocks with diverse mechanical properties. Importantly, the model maintains high performance even when trained only on low-to-mid σ 2 data and tested on high σ 2 conditions—demonstrating strong generalization under extrapolation. This capability has received relatively limited attention in data-driven models and is particularly valuable in practical scenarios where high- σ 2 test data are limited or difficult to obtain. Furthermore, ablation analysis demonstrates that removing physical constraints leads to a notable decrease in model accuracy, underscoring the importance of incorporating strength variation characteristics into the model. SDSNN shows particular advantage under boundary stress conditions and when facing noisy or sparse datasets, indicating its potential to serve as a robust and interpretable tool for true triaxial strength prediction in geotechnical applications.}, archive = {J_EAAI}, author = {Tianzhi Yao and Yunpeng Gao and Jianhai Zhang and Ru Zhang and Li Qian and Qijun Hu and Xianliang Wang and Feng Jiang}, doi = {10.1016/j.engappai.2025.112057}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112057}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A stress-dependent strength neural network model for predicting the true triaxial strength of rocks}, volume = {161}, year = {2025}, }
• (2025). Multi-source contrastive cluster center method for cross-domain bearing fault identification. EAAI, 161, 112056. (www)

  @article{ , abstract = {Because of the complexity and time-varying attributes of the exterior surroundings, rolling bearings commonly operate under variable working conditions at any time. Therefore, there is a multi-source domain adaptation problem that involves multiple source domains and a target domain. In this circumstance, identifying faults directly in a multi-source domain through a model built in a single source domain will also lead to limited model generalization ability, i.e., due to the incompleteness of the training sample size caused by the singularity of the domain quantity limited access to fault knowledge in a single source domain, the established model may be prone to over-fitting, thus whose generalization ability has been decreased under complex and variable working conditions to a certain extent. Hence, this paper has proposed a Multi-source Contrastive Cluster Center (MS3C) Method for addressing the aforementioned issues. Experimental findings on two datasets have suggested that MS3C has not only considered the domain shifts of the same classes between different source domains and the target domain but also adaptively aligned the feature distributions of the same classses in different source domains, therefore, MS3C has a higher identification rate, a better clustering and classification performance and a superior convergence.}, archive = {J_EAAI}, author = {Pengfei Chen and Lizhen Wu and Rongzhen Zhao and Kongyuan Wei and Yuqiao Zheng and Linfeng Deng and Yongfei Zhang and Mingkuan Shi and Zhuo Chen}, doi = {10.1016/j.engappai.2025.112056}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112056}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-source contrastive cluster center method for cross-domain bearing fault identification}, volume = {161}, year = {2025}, }
• (2025). An enhanced encoder–decoder network for temporomandibular joint segmentation in magnetic resonance images. EAAI, 161, 112054. (www)

  @article{ , abstract = {Accurate segmentation of the condyle, articular disc, and articular eminence from magnetic resonance imaging (MRI) is essential for the diagnosis and treatment planning of temporomandibular joint (TMJ) disorders. However, current deep learning methods for TMJ segmentation face significant challenges: (1) anatomical structures of the TMJ exhibit complex inter-patient variability; (2) structural boundaries are often ambiguous; and (3) large inter-slice spacing in MRI makes it difficult to fully utilize spatial context, limiting the effectiveness of both three-dimensional and two-dimensional approaches. To address these challenges, we propose an edge-enhanced TransUNet-based encoder–decoder framework that effectively integrates local edge features and global contextual information to obtain accurate and robust TMJ segmentation. Specifically, a cross-slice attention transformer (CAT) is introduced to capture inter-slice dependencies, addressing the challenge of large slice spacing and enhancing the utilization of contextual information across slices. Moreover, a feature enhancement module (FEM) is designed to explicitly fuse edge information, facilitating accurate localization where boundaries are blurred or indistinct. In addition, an attention gate (AG) mechanism adaptively highlights salient anatomical structures and suppresses irrelevant background, improving the model’s overall focus and resistance to noise. Evaluated on the private TMJ MRI dataset, our method achieves Dice similarity coefficients (DSC) of 0.922, 0.834, and 0.837 for the condyle, articular disc, and articular eminence, respectively, outperforming baseline and comparison methods. Further validation on a public knee MRI dataset demonstrates the good generalizability of the proposed method. These results demonstrate robust and precise TMJ segmentation, supporting reliable automated analysis in clinical settings.}, archive = {J_EAAI}, author = {Yilin Hu and Yunan Zhang and Wei Tang and Jixiang Guo}, doi = {10.1016/j.engappai.2025.112054}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112054}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced encoder–decoder network for temporomandibular joint segmentation in magnetic resonance images}, volume = {161}, year = {2025}, }
• (2025). Three-way decision method integrating probabilistic linguistic term sets and prospect theory in large-scale scenarios. EAAI, 161, 112053. (www)

  @article{ , abstract = {Probability linguistic term sets (PLTSs), owing to their ability to effectively represent the uncertainty and complexity of decision information, have been widely adopted in large-scale multiattribute decision-making (MADM) scenarios. However, existing decision-making methods in the PLTS environment typically suffer from information loss in similarity computation, subjectivity in weight determination, and inadequate consideration of decision-makers’ behavioral psychology, which limits their applicability in real-world complex situations. To address these issues, this paper proposes an innovative hybrid three-way decision (TWD) method that integrates prospect theory with multiobjective optimization on the basis of the ratio analysis (MOORA) approach. Compared with existing studies, the main contributions of this paper include (1) an objective weight determination mechanism based on the information entropy of PLTSs, which improves the scientific validity and computational efficiency of weight allocation compared with subjective or complex traditional methods (2) a novel direct similarity degree based on the probability linguistic Jensen–Shannon (PLJS) divergence, which effectively overcomes the information loss caused by indirect distance measures and achieves accurate similarity characterization between PLTSs (3) a new conditional probability determination strategy utilizing θ -level similarity classes that enhances the dynamic adaptability and accuracy of conditional probability estimation and (4) the integration of prospect theory and the MOORA method within the TWD framework, enabling hierarchical classification of alternatives while fully reflecting decision-makers’ psychological and behavioral characteristics, thereby facilitating the classification and ranking of alternatives. Furthermore, through comparative analysis with large-scale real-world air quality assessment data and mainstream PLTS-based decision-making methods, the effectiveness and rationality of the proposed method are comprehensively verified, demonstrating its superiority, robustness, and practical application value.}, archive = {J_EAAI}, author = {Zhaxi Pahua and Haidong Zhang and Yizhu Cairang and Yanping He}, doi = {10.1016/j.engappai.2025.112053}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112053}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-way decision method integrating probabilistic linguistic term sets and prospect theory in large-scale scenarios}, volume = {161}, year = {2025}, }
• (2025). Adaptive feature-level fusion of manifold and deep learning for robust multi-view face recognition. EAAI, 161, 112052. (www)

  @article{ , abstract = {Robust face recognition across varying poses, lighting conditions, and viewpoints remains a significant challenge in engineering applications such as surveillance, authentication, and human–computer interaction. This paper presents Dynamic Manifold–Deep Learning Fusion (DM-DLF), an artificial intelligence (AI)-based framework that integrates manifold learning with deep neural networks using an adaptive feature-level weighting strategy. By combining global semantic features and local geometric structures, the proposed AI model enhances identity recognition under real-world multi-view scenarios. Experimental results show that DM-DLF surpasses recent transformer-based and graph-based models, achieving up to 94.00 % classification accuracy. The method also reduces prediction error and improves training efficiency. These findings confirm DM-DLF as a powerful AI solution for multi-view face recognition in engineering systems where labeled data is limited and view diversity is high.}, archive = {J_EAAI}, author = {Faraein Aeini}, doi = {10.1016/j.engappai.2025.112052}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112052}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive feature-level fusion of manifold and deep learning for robust multi-view face recognition}, volume = {161}, year = {2025}, }
• (2025). Enhancing long-term load forecasting with convolutional informer-based hybrid model. EAAI, 161, 112051. (www)

  @article{ , abstract = {Long-term load forecasting (LTLF) is essential for energy management but challenged by the complexity of non-stationary time series. The Informer model struggles to capture localized peak–valley patterns, while Variational Mode Decomposition (VMD) faces issues with feature complexity. This study proposes a hybrid framework integrating VMD, Informer, and a Convolutional Long Short-Term Memory (CNN-LSTM) module for accurate LTLF. VMD decomposes non-stationary load data into multi-scale intrinsic mode functions, refined through spectral and autocorrelation analyses to ensure robust feature extraction. The Informer employs sparse self-attention for efficient long-sequence modeling, with CNN-LSTM enhancing the decoder to capture localized temporal dynamics. Experiments on non-stationary load time series across multiple prediction horizons demonstrate that the proposed framework significantly improves forecasting accuracy and robustness compared to baseline models, including Informer and its derivatives. By excelling in complex load pattern prediction, the framework supports efficient grid scheduling and resource optimization in energy systems.}, archive = {J_EAAI}, author = {Bin Sun and Xudong Chen and Tao Shen and Liyao Ma}, doi = {10.1016/j.engappai.2025.112051}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112051}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing long-term load forecasting with convolutional informer-based hybrid model}, volume = {161}, year = {2025}, }
• (2025). Fault diagnosis of high-voltage three-phase asynchronous motors using residual neural networks and bidirectional gated recurrent units. EAAI, 161, 112049. (www)

  @article{ , abstract = {High-voltage three-phase asynchronous motors are extensively used in industrial production; however, rotor bar breakage faults are characterized by strong concealment and severe consequences, making early and accurate diagnosis difficult with traditional methods. To address this issue, this paper proposes a deep learning model based on a Residual Neural Network (ResNet) and a Bidirectional Gated Recurrent Unit (Bi-GRU) for motor fault diagnosis. The proposed model integrates the spatial feature extraction advantages of ResNet with the temporal modeling capabilities of Bi-GRU, simultaneously capturing spatiotemporal information from voltage and current signals. The model was trained and tested on a large-scale dataset comprising 324,000 data points under nine distinct operating conditions. Experimental results demonstrate that the proposed method achieves over 99.83 % in key metrics such as accuracy, precision, recall, and F1-score, significantly outperforming traditional approaches and other comparative models. Additionally, it maintains stable performance under varying operating conditions, demonstrating strong robustness and generalization capability. Further ablation experiments also validated the effectiveness of each module within the proposed model. This study indicates that the application of deep learning in industrial motor fault diagnosis has promising prospects and practical value.}, archive = {J_EAAI}, author = {Huihui Yang and Yuxin Wu}, doi = {10.1016/j.engappai.2025.112049}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112049}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of high-voltage three-phase asynchronous motors using residual neural networks and bidirectional gated recurrent units}, volume = {161}, year = {2025}, }
• (2025). Zero-shot image translation via query compensation and style enhancement. EAAI, 161, 112048. (www)

  @article{ , abstract = {In zero-shot image translation methods based on diffusion models, two prevalent challenges are the loss of identity information and insufficient stylization. To address these issues, we propose a zero-shot image translation method. Specifically, we use the edit-friendly noise space for image inversion. By this way, the identity of the input image is largely preserved. Additionally, we introduce an identity compensation mechanism by injecting the source query vectors into the denoising process. Furthermore, to tackle the model’s insufficient stylization ability, we propose cross-attention style modulation (CSM) to transfer style information from the reference to the input image. Finally, to further enhance the style effect of the translated image, we design initial latent self-enhancement (ILS), style supplementation (SS), and style alignment (SA) strategies. Our zero-shot image translation method does not require training samples, optimization, or fine-tuning. We achieve this by manipulating the self-attention features of a pre-trained diffusion model in a manner analogous to cross-attention—by replacing the key and value vectors of the input image with those of the reference image during the denoising process. Extensive in-domain and cross-domain translation experiments demonstrate the effectiveness of our method across a wide range of object categories and show strong robustness to variations in object shape, size, posture and instance between the input and reference image. The proposed method achieves competitive performance in identity preservation, as measured by the mean Intersection over Union (mIoU), and attains the best style transfer in terms of Single Image Fréchet Inception Distance (SIFID). In addition, qualitative comparisons demonstrate that our approach significantly outperforms the baselines.}, archive = {J_EAAI}, author = {Heng Zhang and Yi-Jun Yang and Wei Zeng}, doi = {10.1016/j.engappai.2025.112048}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112048}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Zero-shot image translation via query compensation and style enhancement}, volume = {161}, year = {2025}, }
• (2025). Camouflaged object detection with boundary localization in complex backgrounds. EAAI, 161, 112047. (www)

  @article{ , abstract = {The primary challenge of Camouflaged Object Detection (COD) lies in the high similarity between the target and the complex background, making it difficult for the human eye to distinguish them. Based on the phenomenon that human attention shifts between the target and the background when observing objects, we propose a network model named MENet. This model adopts a three-stage decoupled architecture of “localization-interaction-fusion.” In the localization stage, we utilize an attention mechanism-based backbone network (Pyramid Vision Transformer V2, abbreviated as PVT-V2) to generate multi-level features, which can initially locate the target area. In the interaction stage, we design a Contour-Aware Edge Module (CAEM) and an Area Decoder (AD) to capture the target edges and background information, respectively, thereby achieving precise localization of the target boundary and reducing interference from background noise. Furthermore, we developed a Boundary Guidance Module (BGM) that effectively injects boundary cues and relevant background information separately into the multi-level features, enhancing the model’s ability to detect target edges in complex backgrounds. In the fusion stage, we design two Feature Fusion Modules (FFM and KFFM) to effectively merge multi-level features with precise boundaries and de-noised features, thereby enhancing the prediction performance of camouflaged objects. Extensive experiments on three challenging benchmark datasets demonstrate that our MENet outperforms many existing state-of-the-art methods. Our method leverages artificial intelligence (AI) techniques to improve the accuracy of camouflaged object and pest detection in complex visual environments. Our code is publicly available at: https://github.com/yang19950966666/MENet .}, archive = {J_EAAI}, author = {Guangjian Zhang and Zhengming Yang and Yong Wang and Yuliang Chen and Duoqian Miao}, doi = {10.1016/j.engappai.2025.112047}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112047}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Camouflaged object detection with boundary localization in complex backgrounds}, volume = {161}, year = {2025}, }
• (2025). A grid-based boundary sharpening clustering algorithm. EAAI, 161, 112045. (www)

  @article{ , abstract = {To address the clustering problem for arbitrary shapes, in this paper, we propose a Grid-based Boundary Sharpening clustering algorithm called as “GBSharp”. This method is grounded in morphology and relies on two fundamental morphological operations: dilation and erosion. The main innovations of the proposed algorithm lie in two aspects. Firstly, we further introduce the concepts of inward dilation and bridge erosion based on the basic morphological operations to reduce the impact of the chain effect. Secondly, a unique indexing structure is designed specifically for non-empty cells in high dimensional space. In addition, to tackle the complex conditional judgments encountered in high-dimensional scenarios, we further utilize the inversion method for bridge-erosion operation. Experiments conducted on synthetic datasets and real-world datasets further validate the effectiveness and efficiency of the proposed algorithm.}, archive = {J_EAAI}, author = {Lin Ma and Qijing Yan and Mengxia Lv and Tiefeng Ma and Mingchang Cheng}, doi = {10.1016/j.engappai.2025.112045}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112045}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A grid-based boundary sharpening clustering algorithm}, volume = {161}, year = {2025}, }
• (2025). Advances in physics-informed neural networks for solving complex partial differential equations and their engineering applications: A systematic review. EAAI, 161, 112044. (www)

  @article{ , abstract = {Most physical and engineering problems can be described by partial differential equations (PDEs), which are typically solved using numerical methods such as the finite difference method and the finite element method. However, conventional numerical discretization approaches face significant challenges in terms of computational efficiency and convergence speed when dealing with complex nonlinear PDEs, such as high-dimensional nonlinear PDEs, stiff PDEs, PDEs with complex boundary conditions or irregular geometries, and multi-scale PDEs. Recently, physics-informed neural networks (PINNs) have emerged as a transformative methodology for solving complex PDEs by integrating physical laws intrinsically into deep learning architectures. While PINNs effectively overcome mesh dependency and dimensionality constraints inherent in traditional numerical methods, they still encounter persistent challenges related to training convergence and generalization robustness. This paper aims to present a comprehensive review of the state-of-the-art developments in PINNs for solving complex PDE problems. The core ideas, network architectures, and generic implementation frameworks, along with associated open-source Python libraries, are first introduced in detail. Furthermore, a systematic taxonomy of optimization techniques is provided, covering hyperparameter selection, adaptive sampling strategies, physics-constrained loss formulations, hybrid differentiation approaches, and architectural innovations. Subsequently, various coping strategies and research advancements of PINNs in addressing complex nonlinear PDE problems are thoroughly discussed. Real-world engineering applications are then reviewed across multiple domains, including cosmology and quantum mechanics, materials science and manufacturing, fluid mechanics, energy systems, biological and environmental sciences, and power and information technologies. Finally, this paper discusses the current challenges and limitations of PINNs in solving complex PDEs and outlines potential directions for future research. By addressing the current limitations and pursuing targeted improvements in architectures, training, interpretability and generalization, PINNs can become a powerful tool in engineering and scientific applications.}, archive = {J_EAAI}, author = {Jiangtao Guo and Hao Zhu and Yujie Yang and Chenrui Guo}, doi = {10.1016/j.engappai.2025.112044}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112044}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advances in physics-informed neural networks for solving complex partial differential equations and their engineering applications: A systematic review}, volume = {161}, year = {2025}, }
• (2025). DPMF-net: A dual-path perceptive multi-stage fusion network for skin lesion segmentation. EAAI, 161, 112043. (www)

  @article{ , abstract = {Accurate segmentation of skin lesions in dermoscopic images is crucial for skin cancer detection and treatment. Despite progress in deep learning-based methods, challenges remain due to diverse skin lesion shapes, colors, and blurred boundaries. We propose a novel Dual-path Perceptive Multi-stage Fusion Network (DPMF-Net) for skin lesion segmentation. DPMF-Net integrates multiple feature refinement modules. It aims to gradually optimize lesion representations by leveraging the dual-path framework to perceive high-level contextual information. The Spatial Frequency Dual-path Cascaded Perception Module (SFDCP) synergizes spatial and frequency domains to model long-range dependencies and suppress noise, enhancing perception of low-contrast lesions. Subsequent to the SFDCP, the Spatial Channel Dual-path Parallel Perception Module (SCDPP) employs entropy-driven attention and multi-granularity convolutions in skip connections to dynamically select informative channels and extract lesion details across spatial scales. To verify the efficacy of our proposed DPMF-Net, extensive experimental assessments are carried out across four challenging datasets. The outcomes of these quantitative and qualitative experiments confirm that our approach significantly outperforms current state-of-the-art methods in terms of all evaluation metrics.}, archive = {J_EAAI}, author = {Yuling Huang and Yaoyao Ma and Jing Wang and Chao Xu and Zhiwei Fan and Di Wu}, doi = {10.1016/j.engappai.2025.112043}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112043}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DPMF-net: A dual-path perceptive multi-stage fusion network for skin lesion segmentation}, volume = {161}, year = {2025}, }
• (2025). Gradient surgery: A necessity for robust test-time adaptation for detecting casting defects. EAAI, 161, 112039. (www)

  @article{ , abstract = {Casting defects pose a significant challenge in the manufacturing industry, leading to material waste, production inefficiencies, and compromised product quality. While deep learning models have shown promise in automating defect detection, their effectiveness is often constrained by domain shifts and variability in real-world data distributions. In this work, we propose Bayesian Test-Time Adaptation (BTTA), a novel framework designed to enhance the robustness and adaptability of machine learning models in such dynamic environments. Unlike traditional Test-Time Adaptation (TTA) methods, our approach employs gradient-guided diversification with Stein Variational Gradient Descent (SVGD) to explore diverse optimization paths. Experimental results on benchmark datasets, including CIFAR-10-C , Casting Defects , and GDXray , demonstrate significant performance improvements across key metrics. Notably, the framework achieves an average accuracy improvement of 2-3% under severe corruption levels and excels in cross-domain generalization, highlighting its ability to handle diverse and unseen defect categories. This dynamic adaptability not only addresses the limitations of static models but also offers a practical and cost-effective solution for real-time defect detection in industrial settings. Our study underscores the potential of BTTA to transform quality assurance processes, ensuring reliable performance across varying operational conditions without the need for extensive retraining or large annotated datasets. The codebase for BTTA is available on: https://github.com/afsharshamsi/GradSurgery .}, archive = {J_EAAI}, author = {Afshar Shamsi and Rejisa Becirovic and Hamid Alinejad-Rokny and Arash Mohammadi and Ahmadreza Argha}, doi = {10.1016/j.engappai.2025.112039}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112039}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gradient surgery: A necessity for robust test-time adaptation for detecting casting defects}, volume = {161}, year = {2025}, }
• (2025). Enhancing real-time detection transformer for small floating targets. EAAI, 161, 112038. (www)

  @article{ , abstract = {Addressing the challenge of detecting small floating target in complex water environments, where it is difficult to balance real-time performance and end-to-end capabilities, this study introduces a specialized model called the Enhancing Real-Time Detection Transformer (ERT-DETR). This model integrates a Dynamic Feature Pyramid Network (DFPNet) with a Micro-Attention Module (MAM) to achieve refined feature extraction and enhanced object recognition mechanisms. Additionally, it incorporates an Inner Intersection over Union (Inner-IoU) auxiliary bounding box loss function, which accelerates model convergence and improves bounding box accuracy. Experiments conducted on the Floating Object in Water Image Dataset (FloW-Img) demonstrate that the ERT-DETR model achieves a 92.6% Average Precision (AP) and 118.84 Frames Per Second (FPS), outperforming the baseline Real-Time Detection Transformer (RT-DETR) by 4.3% and 19.8%, respectively. The ERT-DETR model’s precision and real-time performance in dynamic water surface environments surpass those of the most advanced You Only Look Once (YOLO) and Detection Transformer (DETR) series detectors. This model is significant for enhancing small floating target detection capabilities in complex water environments and can be extended to applications in marine management, environmental monitoring, and vessel tracking.}, archive = {J_EAAI}, author = {Guobing Xie and Xinran Wu and Jiefeng Shi and Yixin Su and Binghua Shi}, doi = {10.1016/j.engappai.2025.112038}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112038}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing real-time detection transformer for small floating targets}, volume = {161}, year = {2025}, }
• (2025). Non-contact weight intelligent estimation based on yak skeleton localization. EAAI, 161, 112036. (www)

  @article{ , abstract = {Weight estimation is a vital method for monitoring the growth and health of yaks, however, traditional techniques-such as relying on herders’ experience or using weighbridge-are labor-intensive, time-consuming and pose safety risks. Currently, many studies have shown that yak body size can be an effective indicator of weight. With the advancement of computer vision, non-contact weight estimation has become increasingly feasible for livestock. Yet, studies focusing on yak weight estimation, particularly in high-altitude plateau regions, remain limited. Herein, we propose a novel weight estimation approach based on deep learning and binocular vision technology to address this gap. The method involves four main steps: (1) yak image acquisition, (2) skeletal key point localization, (3) body size calculation (4) weight estimation using Gaussian process regression. To enhance practicality and mobility, we also developed two edge-intelligent devices: an intelligent inspection vehicle and a handheld detection unit, enabling convenient and non-invasive weight estimation. Our models are trained and tested on a yak dataset collected by our team on the Tibetan Plateau. Experimental results demonstrate the effectiveness of our approach, achieving an Mean Absolute Percentage Error (MAPE) of 0.12 percent, a Mean Absolute Error (MAE) of 25.4 kilograms (kg) and a Coefficient of determination ( R 2 ) value of 0.72. It not only provides a new technical solution for the yak industry but also provides innovative insights for advancing intelligent animal husbandry. The code and data can be accessed at https://github.com/FeiWang-swun/YakWeight .}, archive = {J_EAAI}, author = {Fei Wang and Xinghua Zou and Zhijiang Chen and Qi Tang and Tianshuo Li and Shuiying Wang and Lijun Yang and Dongming Tang}, doi = {10.1016/j.engappai.2025.112036}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112036}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Non-contact weight intelligent estimation based on yak skeleton localization}, volume = {161}, year = {2025}, }
• (2025). Comparison of the performances of artificial intelligence bots using continuous intuitionistic fuzzy evaluation based on distance from average solution method. EAAI, 161, 112033. (www)

  @article{ , abstract = {The rapid evolution of artificial intelligence (AI) has introduced novel opportunities and challenges in various fields. In this study, we present a pioneering approach known as Continuous Intuitionistic Fuzzy (CINFU) Evaluation based on Distance from Average Solution (EDAS), an innovative extension of the EDAS method tailored to Continuous Intuitionistic Fuzzy Sets. This methodology is designed to compare the performance of AI tools. The capabilities of AI bots have been examined through their success rates in various tasks and uncertainty levels in decision-making processes. The study aims to evaluate the effectiveness of different models in decision-making processes by analyzing the performances of AI bots such as Chat Generative Pre-trained Transformer (ChatGPT), Bard, and Claude based on both objective measurements and fuzzy evaluation criteria. The comparison focuses on key performance criteria such as Bots Triggered, User Engagement, Message Click-Through Rate, Chat Handoff, User Retention, Bounce Rate & Dwell Time, Leads Captured, and Customer Satisfaction Score. Ultimately, the validity and robustness of the approach have been tested with sensitivity analysis.}, archive = {J_EAAI}, author = {Nurşah Alkan and Umut Aydın and Akın Menekşe and Cengiz Kahraman}, doi = {10.1016/j.engappai.2025.112033}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112033}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparison of the performances of artificial intelligence bots using continuous intuitionistic fuzzy evaluation based on distance from average solution method}, volume = {161}, year = {2025}, }
• (2025). A global linear attention incorporated video transformer for robust sintering condition recognition. EAAI, 161, 112032. (www)

  @article{ , abstract = {Robust and accurate sintering condition recognition is a fundamental yet critical issue in the design of image-based intelligent combustion control systems. However, owing to the weak texture and fast changing characteristics of flame videos, capturing the condition indicator using existing gradient-based methods is challenging. To address this issue, we propose a global linear attention incorporated video transformer model for sintering condition recognition. First, to reduce the prediction error and uncertainty, the spatial-temporal features are extracted to describe the dynamic characteristics of the flame video streams based on the video shifted window (Swin) Transformer architecture. Next, to address the problem that the local attention strategy used in the Video Swin Transformer is insufficient for global flame feature extraction, we propose a Video Linear Attention block that obtains the global attention as a supplement. Extensive experiments conducted on a real-world rotary kiln sintering dataset demonstrate the effectiveness of our approach, achieving an overall accuracy of 97.76% and an F1-score of 95.30%. Compared to the Video Swin Transformer model, these results represent improvements of 2.00% in accuracy and 4.96% in F1-score, respectively. This research is particularly significant in the context of real-time identification of combustion process conditions, optimization of control parameters, and realization of more stable and efficient combustion process control.}, archive = {J_EAAI}, author = {Leyuan Wu and Junlin Wu and Dingxiang Wang and Qiang Fu}, doi = {10.1016/j.engappai.2025.112032}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112032}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A global linear attention incorporated video transformer for robust sintering condition recognition}, volume = {161}, year = {2025}, }
• (2025). Markerless gait analysis for parkinson’s disease diagnosis: A study on machine learning integration and features explainability. EAAI, 161, 112031. (www)

  @article{ , abstract = {Parkinson’s disease is a common neurodegenerative disorder with progressive loss of dopaminergic and other subcortical neurons that significantly impair motor functions, necessitating accurate diagnostic and monitoring techniques. Researchers commonly use gait analysis to estimate gait parameters to classify diseases and their progression. Traditional gait analysis, while effective, typically requires expensive, specialized equipment and lacks the flexibility for widespread clinical use. Furthermore, many existing studies on markeless approaches do not use public datasets, giving comparable and benchmarking results. Additionally, these methods often do not incorporate explainability, leaving a gap in understanding which specific gait features most significantly impact diagnostic outcomes. To address this issue, we propose a novel system to support diagnosis and analysis of the gait of Parkinson’s disease. The two-step approach involves pose extraction with vision-based pose estimation and classification employing machine learning techniques based on the subject’s temporal, kinematic, and symmetric characteristics. The explainability technique used in this study shows the important role of features such as the knee angle, cadence, and double support duration in influencing diagnostic outcomes. The system shows promising results for non-invasive, low-cost Parkinson’s disease diagnostics, reaching an accuracy of 0.95, a sensitivity of 0.90, a specificity of 0.97, a Matthew’s Correlation Coefficient (MCC) of 0.88, an Area Under the Curve (AUC) of 0.99.}, archive = {J_EAAI}, author = {Cesare Davide Pace and Alessandro Marco De Nunzio and Claudio De Stefano and Francesco Fontanella and Mario Molinara}, doi = {10.1016/j.engappai.2025.112031}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112031}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Markerless gait analysis for parkinson’s disease diagnosis: A study on machine learning integration and features explainability}, volume = {161}, year = {2025}, }
• (2025). Offshore wind power multi-step forecasting based on multi-scale attention mechanism fusion network. EAAI, 161, 112026. (www)

  @article{ , abstract = {The nonlinear, unstable, and multi-scale characteristics of offshore wind power present significant challenges for grid scheduling and power forecasting. To address this issue, this paper proposes a multi-step forecasting hybrid model for offshore wind power based on the sequence-to-sequence (Seq2Seq) architecture, which is named Multi-scale, Denoising, Attention Fusion Network (mDAFNet). The model integrates a multi-level learnable wavelet decomposition network, frequency-domain interpolation denoising technique, and dual-attention mechanism, and improves forecasting performance through joint high- and low-frequency dual-branch predictions. The process begins by improving data quality through outlier cleaning and feature selection based on grey correlation analysis and mutual information. Subsequently, a multi-level learnable wavelet decomposition network is introduced to extract high- and low-frequency subsequences from the frequency domain and optimize parameters dynamically through backpropagation. To address the noise impact on high-frequency subsequences, frequency domain interpolation denoising technique is applied for noise reduction. Then, a sequence-to-sequence model based on the dual-attention mechanism is used to forecast the denoised high-frequency subsequences, while a Long Short-Term Memory (LSTM) neural network predicts the low-frequency subsequences. Finally, the dual-branch forecasting network enables joint modeling of high- and low-frequency subsequences. We conducted experiments on two offshore wind power datasets. The results indicate that, compared to baseline models, the proposed mDAFNet reduces Mean Absolute Error, Root Mean Square Error, and Mean Absolute Percentage Error by an average of 33.94 %, 36.62 %, and 39.84 %, respectively, while significantly improving forecasting accuracy and stability. This study provides a new approach and perspective for offshore wind power forecasting.}, archive = {J_EAAI}, author = {Yizhuang Xiong and Wenbo Wang}, doi = {10.1016/j.engappai.2025.112026}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112026}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Offshore wind power multi-step forecasting based on multi-scale attention mechanism fusion network}, volume = {161}, year = {2025}, }
• (2025). Weighted multi-granularity fuzzy probabilistic rough set based on semi-overlapping function and its application in three-way decision. EAAI, 161, 112023. (www)

  @article{ , abstract = {This paper introduces a novel model of fuzzy probabilistic rough sets and develops an interesting three-way decision (TWD) applicable to multi-attribute group decision-making and classification problems. First, we simultaneously consider the cut-levels of fuzzy relations and decision-making fuzzy sets, investigating a new fuzzy probabilistic rough set model. This model employs semi-overlapping functions as the range of the membership functions of the fuzzy set, enhancing interpretability since the semi-overlapping functions can act as the truth tables of fuzzy logic. Second, by assigning different weights to various granular fuzzy relations, we propose a weighted multi-granularity fuzzy probabilistic rough set model. It is proven that the proposed (weighted multi-granularity) fuzzy probabilistic rough set model encompasses many existing models as special cases, thereby providing a very broad framework for rough set theory. Third, by integrating the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method with Bayesian theory, we develop a method for computing the threshold of fuzzy probabilistic rough sets and construct a new TWD. Sorting and classification rules and algorithms for TWD are provided. Finally, we perform parameter analysis and comparative experiments on nine public datasets to verify the effectiveness and rationality of our TWD. The experimental results demonstrate that our approach exhibits superior fault tolerance and noise reduction capabilities. Specifically, the introduction of semi-overlapping functions enhances our TWD model’s performance, with A c c u r a c y improving the most on the Heart dataset, reaching 9.37%, and the F 1 score increasing the most on the Heart and Plrx datasets, reaching 5.3%. In summary, this paper offers a novel solution for applying artificial intelligence in decision-making scenarios.}, archive = {J_EAAI}, author = {Xinru Li and Lingqiang Li and Chengzhao Jia}, doi = {10.1016/j.engappai.2025.112023}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112023}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted multi-granularity fuzzy probabilistic rough set based on semi-overlapping function and its application in three-way decision}, volume = {161}, year = {2025}, }
• (2025). Feature-driven double deep Q-network with iterated greedy for intelligent scheduling optimization in reentrant hybrid flow shops. EAAI, 161, 112012. (www)

  @article{ , abstract = {The reentrant hybrid flow shop scheduling problem (RHFSP) poses significant challenges due to its complex structure and limited research on model accuracy, solution space exploration, and adaptive strategy selection. To address these gaps, this paper proposes a novel feature-driven double deep Q-network with iterated greedy (FD3QNIG) algorithm. First, a five-dimensional mixed-integer linear programming (MILP) model is developed to significantly improve modeling accuracy. Second, a feature-driven initialization strategy (FDNEH) enhances the quality of initial solutions, while an adaptive historical information-driven destruction and reconstruction strategy (AHDDR) effectively balances exploration and exploitation during the search process. Third, multi-scale local search strategies are employed, including the first application of comprehensive exploration-driven mandatory operations local search (CED_MOLS) to RHFSP, substantially deepening the search capability. Fourth, an adaptive strategy selection mechanism based on double deep Q-Network (DDQN_ASS) dynamically guides the algorithm toward more efficient decision-making. Extensive experiments on 285 benchmark instances demonstrate that FD3QNIG achieves a 53 %–94 % improvement in average relative percentage increase ( A R P I ) over state-of-the-art methods, confirming its effectiveness and robustness.}, archive = {J_EAAI}, author = {Chexiang Li and Yuyan Han and Yuting Wang and Yiping Liu and Biao Zhang and Leilei Meng}, doi = {10.1016/j.engappai.2025.112012}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112012}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-driven double deep Q-network with iterated greedy for intelligent scheduling optimization in reentrant hybrid flow shops}, volume = {161}, year = {2025}, }
• (2025). A semi-supervised framework for generating multi-dimensional taxonomies from asset maintenance documents. EAAI, 161, 112010. (www)

  @article{ , abstract = {The operation and maintenance of buildings generate large volumes of unstructured textual data, such as inspection reports and service requests. These records contain valuable insights that can support fault detection, cost tracking, and resource planning. However, existing classification approaches often rely on static, expert-defined labels that fail to reflect the complexity of real-world maintenance operations. This paper introduces a hybrid framework that combines sentence embedding, clustering, topic modeling, and network modularization to uncover recurring patterns in maintenance text. The extracted patterns are then reviewed and refined by facility management experts to develop a multi-dimensional taxonomy model tailored to operational needs. The methodology is applied to a case study involving over 30,000 work orders. The results demonstrate how the proposed system captures fine-grained details such as system type, failure mode, and required trade expertise. A proof-of-concept software tool, developed in collaboration with facility managers, showcases the practical value of the taxonomy in enabling data-driven decision-making, such as identifying cost drivers and recurring issues. Additionally, the resulting taxonomy models serve as effective prompts for zero-shot text classification, enabling large language models to classify new maintenance records without requiring retraining or labeled data. This approach provides a scalable and adaptable foundation for text classification systems in asset management.}, archive = {J_EAAI}, author = {Soroush Sobhkhiz and Tamer El-Diraby}, doi = {10.1016/j.engappai.2025.112010}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112010}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semi-supervised framework for generating multi-dimensional taxonomies from asset maintenance documents}, volume = {161}, year = {2025}, }
• (2025). A comprehensive review of facial beauty prediction using deep learning techniques. EAAI, 161, 112009. (www)

  @article{ , abstract = {Facial beauty prediction (FBP) is an emerging area of artificial intelligence (AI) that focuses on the development of models that analyze facial features to assess beauty, based on human perception. This task is particularly challenging due to the subjective nature of beauty and limited resources. Deep learning methods have proven their exceptional ability to capture complex features and are therefore well suited for FBP tasks. This paper reviews recent advances in FBP, focusing on deep learning techniques and benchmark datasets. A proposed taxonomy organizes the main methods according to their design and applications, and comparative analyzes highlight trends and the performance of different models. Finally, the study outlines future research directions to drive progress in this evolving field.}, archive = {J_EAAI}, author = {Djamel Eddine Boukhari and Fadi Dornaika and Ali Chemsa and Abdelmalik Taleb-Ahmed}, doi = {10.1016/j.engappai.2025.112009}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112009}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive review of facial beauty prediction using deep learning techniques}, volume = {161}, year = {2025}, }
• (2025). Maximum relevant minimum redundant multi-label feature selection using ant colony optimization. EAAI, 161, 112007. (www)

  @article{ , abstract = {Multi-label learning tasks involve instances that may belong to multiple categories simultaneously, making feature selection particularly challenging in high-dimensional feature spaces. Existing multi-label feature selection methods often suffer from limitations such as high computational complexity, inadequate handling of feature redundancy, and insufficient modelling of label dependencies. To overcome these challenges, we propose a novel framework called Maximum Relevant Minimum Redundant Multi-Label Feature Selection (MR2MLFS), which integrates a two-layer graph representation with a modified Ant Colony Optimization (ACO) strategy. The first graph layer clusters correlated features using Louvain community detection, while the second constructs a meta-graph to model inter-cluster relationships. ACO then explores this structure, favouring the selection of highly relevant and non-redundant features. To reduce computational overhead, we introduce an information-theoretic metric that estimates both feature-label relevance and feature-feature redundancy, eliminating the need for repeated classifier training during the search. We evaluated the proposed method on ten benchmark multi-label datasets using several multi-label classifiers. Experimental results show that the proposed method outperforms six state-of-the-art methods across multiple evaluation metrics, achieving an average relative improvement of 5–12 % while reducing feature dimensionality by up to 80 %. These results confirm the method's robustness, efficiency, and effectiveness in multi-label feature selection.}, archive = {J_EAAI}, author = {Mohammad Hatami and Parham Moradi and Sadegh Sulaimany and Mahdi Jalili}, doi = {10.1016/j.engappai.2025.112007}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112007}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maximum relevant minimum redundant multi-label feature selection using ant colony optimization}, volume = {161}, year = {2025}, }
• (2025). A two-stage model for unified sentence- and document-level biomedical event extraction. EAAI, 161, 112001. (www)

  @article{ , abstract = {Biomedical event extraction, a cornerstone of information extraction, has increasingly attracted attention within the biomedical research community. Moreover, it is a highly complex task, which not only deals with many sub-tasks but also involves nested events. Currently, the research on biomedical event extraction, whether pipelined model or joint method, needs to be processed for each sub-task. The process of processing each sub-task one by one lead to the degradation of event extraction performance. In addition, most studies focus on extracting sentence-level events and ignore cross-sentence event information. To solve these problems, we simplify the process of event extraction, reduce the processing steps, and combine the two sub-tasks of relation extraction and argument combination as one sub-task. In addition, we consider document-level event extraction, which not only extracts cross-sentence events but also considers broader context information. Experimental results indicate that our novel approach outperforms prior studies. Additionally, the document-level event extraction model attains the top performance on the BioNLP’11 test data and achieves near-leading performance on the BioNLP’13 test data.}, archive = {J_EAAI}, author = {Fangfang Su and Yue Zhang and Pengfei Jiao and Zhidong Zhao and Bobo Li and Fei Li and Donghong Ji}, doi = {10.1016/j.engappai.2025.112001}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112001}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A two-stage model for unified sentence- and document-level biomedical event extraction}, volume = {161}, year = {2025}, }
• (2025). A position-aware sets based weakly supervised framework for whole-slide subtype classification. EAAI, 161, 111998. (www)

  @article{ , abstract = {Identifying cancer subtypes is essential for personalized treatment and accurate prognosis due to the varying sensitivities of subtypes to therapies. However, in cancer subtype classification tasks, normal slides are usually scarce or absent as negative samples, while subtype whole-slide images (WSIs) often contain extensive unannotated normal tissue regions. These regions introduce significant noise during feature fusion and subtype classification, leading to degraded performance of existing weakly supervised methods In this paper, we propose the Position-aware Sets based Weakly Supervised learning framework (PSWS), designed for cancer subtype classification using WSIs, with a two-stage structure to enhance model efficiency. Specifically, it first presents a novel patch organization approach, distinct from the bag concept of traditional Multiple Instance Learning (MIL), called position-aware sets, as basic units for learning. Then, PSWS automatically selects subtype-specific features based on enhanced histological features and mutual-patch relations, mitigating the negative impact of unannotated negative regions. In the experiments, the superior performance of PSWS over representative MILs is validated through subtype classification tasks on both public datasets and our internally constructed dataset. Furthermore, class probabilities of position-aware sets and attention region visualizations demonstrate its post-hoc interpretability, assisting pathologists in locating suspicious areas.}, archive = {J_EAAI}, author = {Jiuman Song and Bo Yu and Xiaomin Liu and Lele Cong and Zilong Zhou and Xianling Cong and Hongyan Sun and Shuchao Pang and Hechang Chen}, doi = {10.1016/j.engappai.2025.111998}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111998}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A position-aware sets based weakly supervised framework for whole-slide subtype classification}, volume = {161}, year = {2025}, }
• (2025). Action understanding in low-light and pitch-dark conditions: A comprehensive survey. EAAI, 161, 111996. (www)

  @article{ , abstract = {Action understanding in low-light and dark environments (AULLD) is a critical and technically demanding challenge at the intersection of computer vision and engineering applications. This task focuses on interpreting human actions using motion cues under visually constrained conditions, such as poor illumination or complete darkness. Since 2010, machine learning and, more notably, deep learning have significantly advanced AULLD by enabling systems to extract and learn discriminative features from complex, low-visibility data. These techniques have resulted in state-of-the-art solutions, with recent methods achieving accuracies on various benchmark datasets of up to 97.10%, and a 73.20% top-1 accuracy on an infrared dataset. These solutions have had applications across domains such as healthcare monitoring, surveillance and security, human–computer interaction, and social computing. In this survey, we provide the first comprehensive overview of the progress of the research and topical developments in AULLD. We cover a broad range of elements, including the datasets, evaluation protocols, methods, challenges, and emerging research directions. To organize and evaluate the literature in a structured, integrated way, we introduce a novel taxonomic framework, grounded in four core dimensions: body representation, temporal representation, feature representation, and neural architectures. Employing this taxonomy, we perform an analysis of AULLD approaches, exploring their methodological foundations, system design considerations, performance outcomes, and inherent trade-offs. The survey concludes with an in-depth discussion of the existing challenges confronting AULLD, highlighting several promising research directions that hold potential for advancing the field and setting the stage for future innovations and improvements in AULLD.}, archive = {J_EAAI}, author = {Muhammad Munsif and Samee Ullah Khan and Noman Khan and Altaf Hussain and Sung Wook Baik}, doi = {10.1016/j.engappai.2025.111996}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111996}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Action understanding in low-light and pitch-dark conditions: A comprehensive survey}, volume = {161}, year = {2025}, }
• (2025). Transfer learning framework integrating attention mechanism and domain adaptation for low earth orbit satellite network traffic prediction. EAAI, 161, 111992. (www)

  @article{ , abstract = {Traffic prediction is a crucial prerequisite for planning and even network security in Low Earth Orbit (LEO) satellite networks (LSNs). This paper designed a transfer learning framework for LSN traffic prediction that leverages attention mechanism and domain adaptation. Firstly, by integrating five-dimensional data including global population distribution, local time coefficient, the Internet penetration rate of the country, daily data volume of a single Internet user, and global aeronautical traffic demand, a traffic model that could characterize the traffic situation of the area covered by LEO satellite within a specific time range was constructed. Considering the problem of insufficient online traffic data, knowledge was transferred from terrestrial network traffic (source domain) to satellite network traffic (target domain) by incorporating the Domain-Adversarial Neural Network (DANN) method to tackle the data distribution discrepancies between the source and target domains. Finally, by combining DANN with the attention mechanism, the domain-invariant features of the source domain and the target domain were extracted to predict satellite network traffic accurately. Experimental results show that compared to baseline models, the error of the proposed framework in terms of root mean square error measurement is reduced by 9.57% to 33.47% and 18.85% to 38.99% in the two simulated LSN traffic scenarios. Moreover, this framework has low computational complexity among other transfer learning models, which can lay a foundation for subsequent satellite traffic planning and network security.}, archive = {J_EAAI}, author = {Yan Zhang and Yong Wang and Qingsong Zhao and Yadi Zhai and Zhi Lin and Luda Zhao and Yihua Hu}, doi = {10.1016/j.engappai.2025.111992}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111992}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning framework integrating attention mechanism and domain adaptation for low earth orbit satellite network traffic prediction}, volume = {161}, year = {2025}, }
• (2025). Skeleton-based multi-person action recognition towards real-world violence detection. EAAI, 161, 111987. (www)

  @article{ , abstract = {Human action recognition is crucial in intelligent systems like robotics, healthcare, and surveillance, gaining significant attention with the rise of AI in computer vision. This work presents the Multi-Skeleton Action Recognizer (MSAR), a comprehensive multi-stage process specifically designed to enhance the recognition and classification of violent activities involving multiple individuals. The approach aims to improve accuracy and robustness in real violence detection and classification leveraging skeletal data. The proposed MSAR pipeline consists of five main stages: Human detection, Human tracking, Skeleton extraction, Data pre-processing and Padding, and Action classification. Experimental results demonstrate the effectiveness of our multi-stage process, showing significant improvements in multi-person action recognition in violence detection. In the human detection stage, our integration of multiple detection models provides a comprehensive visual comparison of model performance, offering empirical insights into accuracy, computational efficiency, and processing speed. The fusion Bidirectional Long Short-Term Memory - Gated Recurrent Unit (BiLSTM-GRU) architecture designed for action classification integrating Deep Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (GRU) components, demonstrated maximum accuracy of 96.46% and an average accuracy of 91.18%. The proposed architecture is explored to potentially enhance the model’s ability to capture temporal dependencies in action sequences. Additionally, our proposed multi-stage pipeline for action recognition exemplifies a modular, systematic approach, where each stage functions as an independent module is expanded, optimized, replaced without affecting others. This modularity enhances the flexibility, adaptability, scalability of the pipeline, providing a solid foundation for future advancements across various tasks, application domains.}, archive = {J_EAAI}, author = {Minh-Trieu Truong and Van-Dung Hoang}, doi = {10.1016/j.engappai.2025.111987}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111987}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Skeleton-based multi-person action recognition towards real-world violence detection}, volume = {161}, year = {2025}, }
• (2025). Malware attack and defense game in fractional-order internet of underwater things: Model-based and model-free approaches. EAAI, 161, 111970. (www)

  @article{ , abstract = {With the rapid development of the Internet of Underwater Things (IoUT), its cybersecurity faces increasingly severe challenges, as malware attacks have become a common and highly destructive threat. However, most existing studies rely on integer-order models or static defense mechanisms, which fail to capture the memory and hereditary properties of underwater malware propagation and lack adaptability in dynamic adversarial environments. To address this issue, this paper constructs a fractional-order propagation model and introduces differential game theory to investigate the dynamic interactions between malware and system defense, aiming to derive the optimal dynamic strategies and the corresponding Nash equilibrium for both parties. Furthermore, two innovative artificial intelligence strategy learning approaches are proposed: a model-based method and a model-free method, designed to approximate the optimal control strategies in the game. Extensive comparative experiments and simulation results demonstrate the effectiveness of the proposed methods in strategy learning. We further discuss the practical applications of these methods in IoUT as well as their scalability to other domains.}, archive = {J_EAAI}, author = {Guiyun Liu and Zulong Peng and Tingting Tan and Xiaojing Zhong and Zhongwei Liang}, doi = {10.1016/j.engappai.2025.111970}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111970}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Malware attack and defense game in fractional-order internet of underwater things: Model-based and model-free approaches}, volume = {161}, year = {2025}, }
• (2025). Semi-supervised graph contrastive learning for emotion recognition based on electroencephalogram signals. EAAI, 161, 111969. (www)

  @article{ , abstract = {Emotion recognition plays a crucial role in affective computing, aiming to identify and interpret human emotions. Since electroencephalography (EEG) measures brain electrical activity associated with emotional processing, its use in emotion recognition has garnered significant attention. However, due to the complexity of EEG, capturing the low-dimensional manifold structure from a high-dimensional EEG signal remains a challenge. Moreover, even though EEG labeling is expensive and requires specialization, the development of emotion recognition models has predominantly relied on labeled EEG datasets. To address these issues, we propose a semi-supervised graph contrastive learning (SGCL) model for EEG-based emotion recognition, leveraging the abundance of unlabeled EEG data to generate clearer representations. The proposed SGCL extracts two frequency-domain features from the EEG signal, differential entropy (DE) and power spectral density (PSD), which are integrated through a symmetric similarity network fusion (SSNF) with graph contrastive learning (GCL) to improve the generalization and representation capability of the graph convolutional network (GCN) to complex EEG data in transductive learning tasks. Extensive experimental results on three public datasets reveal that the proposed SGCL consistently outperforms state-of-the-art models. Notably, on the Shanghai Jiao Tong University EEG Emotion dataset (SEED), SGCL achieves an average accuracy of 99.99% using only 3.5% of the data as labeled input, setting a new benchmark in both effectiveness and efficiency. Further analyses confirm our model learns highly discriminative representations and offers insightful explanations, demonstrating its potential for robust emotion recognition.}, archive = {J_EAAI}, author = {Dae Hyeon Kim and Young-Seok Choi}, doi = {10.1016/j.engappai.2025.111969}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111969}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised graph contrastive learning for emotion recognition based on electroencephalogram signals}, volume = {161}, year = {2025}, }
• (2025). Explainable automated wild-orchid identification combining deep neural networks and bayesian networks. EAAI, 161, 111961. (www)

  @article{ , abstract = {Deep learning has been shown repeatedly to be a successful method of obtaining accurate classifiers. This also applies to orchid identification from digital photographs. However, deep neural networks possess the major weakness of lack of explainability, missing the ability to explain the reasons behind a decision. Nevertheless, most current research regarding automated orchid identification applies this blackbox approach. By contrast, in this paper we propose a new method for trustworthy automated orchid identification combining two complementary methods: deep neural networks and feature-based Bayesian networks, where the Bayesian network is also utilized for providing an explanation of the generated solutions. We use other deep neural networks to extract flower characteristics, the features, from the images which are subsequently fed into the Bayesian network as uncertain evidence. When combining the deep neural network and the Bayesian network as an ensemble classifier, both reaching the same conclusion, an accuracy of 89.4% is achieved, the most trustworthy outcome. With a human-in-the-loop ensemble classifier, validation results are even better, yielding an accuracy of 98.1%. Our approach also exploits the taxonomic knowledge represented in the Bayesian network to provide an explanation of the solutions for every case, reinforcing further trust in the method. The result is an explainable user-in-the-loop ensemble classifier. Providing explainability can help build user trust in a system and may play a major role when it is used as a learning aid for new orchid enthusiasts. Finally, the proposed method may be also of value in many fields other than plant determination.}, archive = {J_EAAI}, author = {Diah Harnoni Apriyanti and Luuk J. Spreeuwers and Peter J.F. Lucas}, doi = {10.1016/j.engappai.2025.111961}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111961}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable automated wild-orchid identification combining deep neural networks and bayesian networks}, volume = {161}, year = {2025}, }
• (2025). A safe multi-agent reinforcement learning algorithm using constraint update projection approach. EAAI, 161, 111929. (www)

  @article{ , abstract = {Traditional reinforcement learning has a major limitation, which is that they optimize agent’s policy purely for maximizing rewards. It completely ignore safety considerations. However, in certain critical engineering fields, ensuring safety is of utmost importance, otherwise it can cause incalculable losses. Therefore, this paper proposes a safe Multi-Agent Constrained Update Projection(MACUP) algorithm, which can safely control agents to complete tasks. We solve this problem from the perspective of policy constraint optimization. Firstly, we derive the new bounds of multi-agent policy performance difference based on a tighter general policy performance difference. It contains generalized advantage estimates, and we utilize these bounds as surrogate functions concerning the objective and constraints. Secondly, to address the coordination issue among multiple agents, we employ a multi-agent sequential policy update framework. Finally, we use a projection method to optimize policies, which has low computational complexity and does not require convex approximation of the surrogate function for solving. It can help us reduce errors. Finally, we have validated our algorithm in two different multi-agent safety environments, and the results show that it is able to satisfy safety constraints while achieving higher rewards.}, archive = {J_EAAI}, author = {Yang Liu and Xiang Feng and Huiqun Yu}, doi = {10.1016/j.engappai.2025.111929}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111929}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A safe multi-agent reinforcement learning algorithm using constraint update projection approach}, volume = {161}, year = {2025}, }
• (2025). Multi-criteria group decision-making method using spherical fuzzy Z-numbers for smart technology revolution in municipal waste management. EAAI, 161, 111928. (www)

  @article{ , abstract = {Due to the new phenomenon of globalization and the fast growing cities, there is drastic pressure on the conventional methods of waste management hence the need for innovative management wastage that is proactive to the theme of environmental and economic challenge of the modern world. This paper aims at investigating the following research question: How has the introduction of smart technologies impacted waste management in the context of a mid-sized city that is in the cross-road of generating more wastes while at the same time, concerns over ecological attainability. They present a new type of Spherical Fuzzy Z-Number Sets in organizational environments dealing with multi criteria group decision making where it possess higher order of uncertainty than conventional fuzzy sets. For handling the multi facility nature of multi criteria group decision making in waste management, we propose the so-called Additive Ratio Assessment method when the attribute weights are unknown. To ensure that the criteria weights are determined objectively in this research, the CRITIC (CRiteria Importance Through Intercriteria Correlation) technique is used. The first part of the study provides a theoretical framework of spherical fuzzy Z-numbers concerning accuracy, scoring functions, and operations. Then we introduce this framework to actual multi criteria group decision making cases in municipal waste management to show that how Spherical Fuzzy Z-Number can facilitate decision-making processes by dealing with the vagueness and fuzziness of stakeholder preferences. The TODIM (Tomada de Decisao Iterativa Multicriterio) technique is used in this to validate and compare for efficiency of the proposed additive ratio assessment method. Besides, this research will not only enhance the theoretical aspect of fuzzy decision making but also present a feasible and effective framework for handling unsound and random decision making problems especially within the context of urban waste management.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Muhammad Naeem and Chiranjibe Jana and Maria Akram and Gerhard-Wilhelm Weber}, doi = {10.1016/j.engappai.2025.111928}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111928}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-criteria group decision-making method using spherical fuzzy Z-numbers for smart technology revolution in municipal waste management}, volume = {161}, year = {2025}, }
• (2025). From image processing to artificial intelligence-driven tools: A comprehensive survey on the evolution of feature extraction methods in paintings. EAAI, 161, 111794. (www)

  @article{ , abstract = {This study explores the computational process of converting qualitative elements of paintings, such as shape, color, texture, and line, into quantitative numerical values, which aid in identifying and extracting features from painting images. These identified features are then used to classify paintings based on artist, art style, genre, and art movements. This review employs a systematic literature review methodology, examining approximately 80 research papers to track trends in this field from 2015 to 2025. With the increasing presence of paintings in online media, museums, and galleries, artificial intelligence (AI) plays a significant role in interpreting these subjective elements. This review examines various image processing techniques in conjunction with AI implementations to extract the local and global features of a painting. These methods are combined with AI models, such as deep learning and computer vision algorithms, to improve feature extraction and provide a more thorough and accurate paintings analysis.}, archive = {J_EAAI}, author = {Rekha Sharma and Rishi Gupta and Aditya Sinha}, doi = {10.1016/j.engappai.2025.111794}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111794}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From image processing to artificial intelligence-driven tools: A comprehensive survey on the evolution of feature extraction methods in paintings}, volume = {161}, year = {2025}, }

EJOR - 25

• (2026). Dynamic mode decomposition for online portfolio selection task. EJOR, 328(1), 349-365. (www)

  @article{ , abstract = {Online portfolio selection (OPS) is a complex task aimed at maximizing investment returns through strategic allocation of capital among risky assets. Traditional Follow the Winner (FTW) strategies, grounded in the Best Constant Rebalanced Portfolios strategy, assume market is independent and identically distributed (i.i.d.), which often fails to capture real-world financial market dynamics, leading to sub-optimal performance in practical applications. To address this limitation, we propose integrating Dynamic Mode Decomposition (DMD) into FTW strategies. DMD is a powerful data-driven technique that originated in the field of fluid dynamics. It is designed to extract coherent structures and identify temporal patterns within complex data. By applying DMD to financial market data, we can uncover underlying patterns and trends that are not apparent under the i.i.d. assumption. Significantly, the integrated DMD in this paper allows for efficient recursion, which is particularly crucial for OPS task. To illustrate the effectiveness of the proposed idea, we consider the Exponential Gradient (EG) strategy as an example and proposed Exponential Gradient with Dynamic Mode Decomposition (EGDMD). The results demonstrate that the proposed EGDMD outperforms traditional EG-type strategies, significantly improves risk-adjusted returns, and maintains computational efficiency. The integration of DMD allows for more accurate identification of market patterns, leading to more effective investment decisions and enhanced portfolio performance.}, archive = {J_EJOR}, author = {Jiahao Li and Yong Zhang and Xiaoteng Zheng}, doi = {10.1016/j.ejor.2025.04.049}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {349-365}, shortjournal = {Eur. J. Oper. Res.}, title = {Dynamic mode decomposition for online portfolio selection task}, volume = {328}, year = {2026}, }
• (2026). Choice-based crowdshipping for next-day delivery services: A dynamic task display problem. EJOR, 328(1), 336-348. (www)

  @article{ , abstract = {This paper studies integrating the crowd workforce into next-day home delivery services. In this setting, both crowd drivers and contract drivers collaborate in making deliveries. Crowd drivers have limited capacity and can choose not to deliver if the presented tasks do not align with their preferences. The central question addressed is: How can the platform minimize the total task fulfilment cost, which includes payouts to crowd drivers and additional payouts to contract drivers for delivering the unselected tasks by customizing task displays to crowd drivers? To tackle this problem, we formulate it as a finite-horizon Stochastic Decision Problem, capturing crowd drivers’ utility-driven task preferences, with the option of not choosing a task based on the displayed options. An inherent challenge is approximating the non-constant marginal cost of serving orders not chosen by crowd drivers, which are then assigned to contract drivers. We address this by leveraging a common approximation technique, dividing the service region into zones. Furthermore, we devise a stochastic look-ahead strategy that tackles the curse of dimensionality issues arising in dynamic task display execution and a non-linear (problem specifically concave) boundary condition associated with the cost of hiring contract drivers. In experiments inspired by Singapore’s geography, we demonstrate that choice-based crowd shipping can reduce next-day delivery fulfilment costs by up to 16.9%. The observed cost savings are closely tied to the task display policies and the task choice behaviours of drivers.}, archive = {J_EJOR}, author = {Alp Arslan and Fırat Kılcı and Shih-Fen Cheng and Archan Misra}, doi = {10.1016/j.ejor.2025.05.046}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {336-348}, shortjournal = {Eur. J. Oper. Res.}, title = {Choice-based crowdshipping for next-day delivery services: A dynamic task display problem}, volume = {328}, year = {2026}, }
• (2026). When cost metafrontiers are nonconvex in the outputs, then the production metafrontier is nonconvex: The price of a convexification strategy. EJOR, 328(1), 324-335. (www)

  @article{ , abstract = {Metafrontier analysis is widely used to account for technological heterogeneity among producers. The approach involves combining a number of group-specific production possibilities sets to form a production possibilities metaset. Even though the union of the group sets normally results in a nonconvex metaset, most authors proceed as if this metaset is convex. Kerstens, O’Donnell and Van de Woestyne (2019) obtain new results on the union operator on sets under various assumptions and empirically illustrate that the popular convexification strategy is highly questionable. In this paper we transpose their results on the union operator from a production to a cost context: this is new. We then explore the extent to which convexity of the cost function is corroborated using a newly developed test. Furthermore, we check to which extent a convexification strategy is tenable when estimating a cost metafrontier. We use an original banking data set from China and the USA to illustrate the main issues. We establish that the cost function is not convex in the outputs for China and that the convexification strategy leads to potentially-biased estimates of the cost metafrontier and associated measures of efficiency.}, archive = {J_EJOR}, author = {Kristiaan Kerstens and Christopher O’Donnell and Ignace Van de Woestyne and Shirong Zhao}, doi = {10.1016/j.ejor.2025.05.048}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {324-335}, shortjournal = {Eur. J. Oper. Res.}, title = {When cost metafrontiers are nonconvex in the outputs, then the production metafrontier is nonconvex: The price of a convexification strategy}, volume = {328}, year = {2026}, }
• (2026). Mitigating adversarial attacks on transformer models in credit scoring. EJOR, 328(1), 309-323. (www)

  @article{ , abstract = {The integration of unstructured data, such as text created by borrowers, offers new opportunities for improving credit default prediction but also introduces new risks. This study examines the robustness of transformer-based credit scoring models that utilize textual data and assesses their vulnerability to adversarial attacks. Using peer-to-peer lending data, we show that small, semantically neutral changes in loan descriptions can substantially alter model outputs. These vulnerabilities expose lenders and borrowers to economic risks through distorted risk assessments and mispriced loans. We evaluate two mitigation strategies: adversarial training and topic modeling. Adversarial training improves robustness without compromising predictive performance. Topic modeling provides a more interpretable and stable representation of borrower narratives. An economic analysis confirms that robust models reduce mispricing and improve outcomes for all parties. The findings underscore the importance of robustness as the use of unstructured data in credit scoring becomes more accessible.}, archive = {J_EJOR}, author = {Brandon Schwab and Johannes Kriebel}, doi = {10.1016/j.ejor.2025.05.029}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {309-323}, shortjournal = {Eur. J. Oper. Res.}, title = {Mitigating adversarial attacks on transformer models in credit scoring}, volume = {328}, year = {2026}, }
• (2026). Cost-sensitive single-index classification model. EJOR, 328(1), 295-308. (www)

  @article{ , abstract = {Single-index models (SIMs) are a type of semiparametric model in which a response variable is assumed to be related to a linear combination of explanatory variables by an unknown function, on which any restriction is imposed. Thus, they provide both interpretability and flexibility to capture complex data relationships. In this paper, SIMs are extended to the cost-sensitive classification problem by minimizing the different misclassification costs. The flexibility of SIMs combined with a cost-sensitive approach results in a powerful model to minimize losses and optimize decision making. This is demonstrated through an extensive simulation study and the analysis of five real data sets, where the proposed approach outperforms both parametric and semi-parametric previous approaches.}, archive = {J_EJOR}, author = {Jorge C-Rella and Ricardo Cao and Juan M. Vilar}, doi = {10.1016/j.ejor.2025.08.058}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {295-308}, shortjournal = {Eur. J. Oper. Res.}, title = {Cost-sensitive single-index classification model}, volume = {328}, year = {2026}, }
• (2026). Diverse ensemble cost-sensitive logistic regression. EJOR, 328(1), 282-294. (www)

  @article{ , abstract = {In recent years, cost-sensitive methods have become increasingly crucial for decision-making in various real-world applications. These methods have been developed for the purpose of minimizing costs or risks for stakeholders. Moreover, the interpretability of cost-sensitive methods has gained considerable attention in critical domains such as finance and medical care. In this article, we propose a diverse ensemble of cost-sensitive logistic regression models to reduce costs for binary classification tasks, as well as a novel algorithm based on the partial conservative convex separable quadratic approximation to solve this non-convex optimization problem. The proposed method demonstrates substantial cost savings through extensive simulations and real-world applications, including fraud detection and gene expression analysis. Additionally, unlike other ensembling techniques, the resulting model of the proposed method is fully interpretable as a logistic regression model and achieves a high level of sparsity induced by the proposed algorithm. We believe this approach offers deeper insights into the relationship between predictors and response, enabling more informed decision-making in practical scenarios.}, archive = {J_EJOR}, author = {Bing Yang and Stefan Van Aelst and Tim Verdonck}, doi = {10.1016/j.ejor.2025.07.028}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {282-294}, shortjournal = {Eur. J. Oper. Res.}, title = {Diverse ensemble cost-sensitive logistic regression}, volume = {328}, year = {2026}, }
• (2026). Multi-view ensemble feature selection via SemiDefinite programming. EJOR, 328(1), 269-281. (www)

  @article{ , abstract = {Multi-view learning faces significant challenges in selecting discriminative features while managing redundancy and noise across heterogeneous data sources. To address these issues, this paper introduces Multi-view Ensemble Feature Selection (MEFS), a novel framework that systematically integrates view generation (VG) and view selection (VS) through a unified optimization paradigm. By reformulating feature selection as a MaxCut problem and leveraging SemiDefinite Programming (SDP) relaxation, MEFS dynamically balances the generalization capability of individual views with their pairwise diversity, eliminating the need for manual parameter tuning. A key innovation is the proposed pairwise diversity metric, which quantifies inter-view dissimilarity using between-class scatter matrices to ensure complementary feature subsets. Extensive experiments on ten benchmark datasets demonstrate that MEFS consistently outperforms state-of-the-art methods in accuracy, robustness, and computational efficiency. Ablation studies validate the synergistic effect of combining VG and VS modules.}, archive = {J_EJOR}, author = {Xiaojian Ding and Xin Wang and Pengcheng Shi}, doi = {10.1016/j.ejor.2025.07.014}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {269-281}, shortjournal = {Eur. J. Oper. Res.}, title = {Multi-view ensemble feature selection via SemiDefinite programming}, volume = {328}, year = {2026}, }
• (2026). On the price of diversity for multiwinner elections under (weakly) separable scoring rules. EJOR, 328(1), 258-268. (www)

  @article{ , abstract = {We consider a model of multi-winner elections, where each voter expresses a linear preference over a finite set of alternatives. Based on voters’ preferences, the primary goal is to select a subset of admissible alternatives, forming what is referred to as a committee. We explore (weakly) separable committee scoring rules, the voting mechanisms that assess each alternative individually using a scoring vector and select the top k alternatives, where k represents the committee’s size. Furthermore, we operate under the assumption that alternatives are categorized based on specific attributes. Within each attribute category, there exists a targeted minimum number of alternatives that the selected committee should encompass, emphasizing the necessity for diversity. In this context, we assess the cost associated with imposing such a diversity constraint on the voting process. This assessment is conducted through two methodologies, referred to as the “price of diversity” and the “individual price of diversity”. We set the upper bounds for both prices across all (weakly) separable committee scoring rules. Additionally, we show how the maximum price of diversity can be used to discriminate between different voting rules in this context. Ultimately, we illustrate that concentrating on the candidates’ performance yields a more accurate estimation of the price of diversity compared to a focus on the enforced diversity constraint.}, archive = {J_EJOR}, author = {Mostapha Diss and Clinton Gubong Gassi and Eric Kamwa}, doi = {10.1016/j.ejor.2025.06.013}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {258-268}, shortjournal = {Eur. J. Oper. Res.}, title = {On the price of diversity for multiwinner elections under (weakly) separable scoring rules}, volume = {328}, year = {2026}, }
• (2026). Nonconvex truncated conditional value at risk-based sparse linear regression. EJOR, 328(1), 246-257. (www)

  @article{ , abstract = {Conditional value at risk (CVaR) is a widely recognized risk measure used to manage data uncertainty within risk management. In this paper, we study a class of sparse linear regression models based on truncated CVaR measure and ℓ 0 -norm regularization. Due to the nonconvexity and nonsmoothness of the objective functions, as well as the NP-hardness of the problem with the ℓ 0 -norm regularization, we propose an approximation model that employs a tight relaxation of the ℓ 0 -norm. The solution equivalence between the proposed model and its approximation model is explored. To efficiently solve the approximation model, we develop a semismooth Newton-based proximal majorization-minimization algorithm. Furthermore, the convergence analysis of the proposed algorithm is presented, and the convergence rate for the reduced CVaR-based sparse linear regression model is established. Moreover, extensive numerical experiments conducted on both synthetic and real datasets validate the stability and effectiveness of the proposed algorithm, demonstrating significant improvements in both sparsity and accuracy compared to existing state-of-the-art methods.}, archive = {J_EJOR}, author = {Boyi Xie and Zhongming Wu and Min Li}, doi = {10.1016/j.ejor.2025.06.004}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {246-257}, shortjournal = {Eur. J. Oper. Res.}, title = {Nonconvex truncated conditional value at risk-based sparse linear regression}, volume = {328}, year = {2026}, }
• (2026). Designing and computing explanations for comparisons inferred from an additive value model. EJOR, 328(1), 232-245. (www)

  @article{ , abstract = {Many decision models are based on an additive representation of preferences. Recommendations obtained from such additive decision models are sometimes considered as self-evident. On the contrary, we claim that these recommendations deserve an explanation so as to be fully understood by the user/decision-maker and to foster her trust. We propose to explain a preference statement x preferred to y by decomposing this statement into simpler ones. Arguments in favor of x (Pros), and arguments in favor of y (Cons) are decomposed using a covering scheme in which each Con is covered by a Pro. We use a decomposition language in which elementary self-evident statements involve ( i ) one Pro against one Con, ( ii ) one pro against several Cons, or ( iii ) several Pros against one Con. We prove that computing such explanations is computationally difficult in case ( ii ) and ( iii ), and propose a mathematical programming formulation to solve it. Numerical experiments provide insights on the actual behavior of our algorithm. We also illustrate the usefulness of our approach in the context of multicriteria decision aid but also for machine learning approaches.}, archive = {J_EJOR}, author = {Manuel Amoussou and Khaled Belahcene and Nicolas Maudet and Vincent Mousseau and Wassila Ouerdane}, doi = {10.1016/j.ejor.2025.05.058}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {232-245}, shortjournal = {Eur. J. Oper. Res.}, title = {Designing and computing explanations for comparisons inferred from an additive value model}, volume = {328}, year = {2026}, }
• (2026). Bayesian variable selection in kriging metamodeling for quality design. EJOR, 328(1), 216-231. (www)

  @article{ , abstract = {The increasing complexity of production processes and rapid developments in digital technology have fueled the adoption of metamodels in quality design. Kriging has emerged as one of the most popular emulation methods for both deterministic and stochastic simulations. Conventional Kriging models with predetermined mean functions, such as ordinary or universal Kriging, may exhibit subpar predictive performance when strong trends exist. This paper proposes a novel variable selection procedure for the mean function that ensures prediction accuracy while using only a limited number of variables to capture the potential existing trends in deterministic simulations. The proposed method integrates the benefits of Bayesian variable selection and frequentist statistical tests. Initially, a group of potential models is chosen to build the mean function, employing the Bayesian method with priors designed to guarantee sparsity. This results in a significant reduction in the number of models to be considered in the next stage. Subsequently, each candidate model undergoes rigorous frequentist tests to thoroughly assess its reliability and validity. Extensive simulation studies are conducted using the well-known Borehole function and a real-life case. The results demonstrate the superiority of the proposed method over several existing approaches, establishing its effectiveness in achieving robust parameter design.}, archive = {J_EJOR}, author = {Baoping Tao and Zifei Han and Wen Shi and Min Wang and Linhan Ouyang}, doi = {10.1016/j.ejor.2025.06.003}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {216-231}, shortjournal = {Eur. J. Oper. Res.}, title = {Bayesian variable selection in kriging metamodeling for quality design}, volume = {328}, year = {2026}, }
• (2026). Design and pricing of extended warranty menus with reference effects. EJOR, 328(1), 201-215. (www)

  @article{ , abstract = {Consumer durables increasingly come with extended warranty menus—beyond manufacturers’ base warranties—that offer multiple options with differentiated protection lengths and prices. When choosing from an extended warranty menu, consumers might not only evaluate the intrinsic utility of each option but also form a reference point against which the available options are compared. In this paper, we posit that consumers compare the price-length ratios of available options when determining their willingness to pay, and investigate the design and pricing of extended warranty menus under such reference effects. To this end, we adapt the standard multinomial logit choice model to incorporate two types of reference point that are generated endogenously and exogenously, respectively, with respect to the given menu. We show that if the warranty options are ordered in the protection length, then the optimal pricing policies prescribe the same adjusted markup and the same price-length ratio alternately. We further extend our analysis to price competition under reference effects, where multiple firms compete in the aftermarket, each offering a single extended warranty. We prove the existence of a unique Nash equilibrium and develop an efficient method to identify it. Numerical examples are presented to illustrate the analytical findings, and sensitivity analyses are conducted to examine the impact of reference-effect coefficients on the optimal pricing policies. Overall, this work highlights the importance of incorporating reference effects into the design and pricing of extended warranty menus.}, archive = {J_EJOR}, author = {Xiao-Lin Wang and Chenglong Li and Junjie Wang}, doi = {10.1016/j.ejor.2025.05.056}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {201-215}, shortjournal = {Eur. J. Oper. Res.}, title = {Design and pricing of extended warranty menus with reference effects}, volume = {328}, year = {2026}, }
• (2026). Planning bayesian reliability demonstration tests via a generalized test statistic. EJOR, 328(1), 189-200. (www)

  @article{ , abstract = {Reliability demonstration testing (RDT) has been extensively employed to verify whether a product meets specific reliability requirements at a desired confidence level. Driven by intense market competition and constrained test resources, manufacturers are motivated to seek effective strategies to reduce the testing efforts required for RDT. In this paper, we propose a method that utilizes existing knowledge and information obtained from the product design and development phase to construct a Bayesian prior distribution of the product’s reliability. Based on this prior, a preliminary disposition decision on whether to accept or reject the product is made. A subsequent demonstration test is needed only when the prior information is deemed insufficient for an immediate disposition. A RDT planning method is developed based on the posterior distribution of the product’s reliability, which is applicable to general cases involving non-conjugate priors. We study two types of demonstration testing: binomial and exponential. For each, we prove the existence of an optimal test plan and develop an efficient searching algorithm to determine it. Numerical studies are conducted to demonstrate the effectiveness of the proposed method, supplemented by a case study on RDT for systems of different configurations. Overall, this work provides a unified and effective framework for reliability demonstration under the Bayesian paradigm.}, archive = {J_EJOR}, author = {Zan Li and Jianyu Xu and Chengjie Wang and Xiao-Lin Wang}, doi = {10.1016/j.ejor.2025.08.011}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {189-200}, shortjournal = {Eur. J. Oper. Res.}, title = {Planning bayesian reliability demonstration tests via a generalized test statistic}, volume = {328}, year = {2026}, }
• (2026). A primal–dual policy iteration algorithm for constrained markov decision processes. EJOR, 328(1), 174-188. (www)

  @article{ , abstract = {The solution algorithms of Constrained Markov Decision Process (CMDP), a widely adopted model for sequential decision-making, have been intensively studied in the literature. Despite increasing effort, the Linear Programming (LP) formulation of CMDP remains the dominant exact method that leads to the optimal solution without constraint violations. However, the LP formulation is computationally inefficient due to the curse of dimensionality in CMDP state and action spaces. In this study, we introduce a novel policy iteration method for CMDP, based on decomposition and row-generation techniques. We design a Primal–Dual Policy Iteration (PDPI) algorithm that utilizes state values and Lagrangian multipliers to improve randomized stationary policies in an iterative fashion. We analytically show that upon convergence, PDPI produces the optimal solution for CMDP. An upper bound of the convergence iterations is also given. To validate the algorithm performance, we conduct comprehensive computational experiments on six benchmarking problems curated from the literature. Results show that PDPI outperforms conventional methods considerably, improving the total algorithm runtime by up to 89.19%. The improvement becomes more significant as the problem size grows larger. We further provide insights and discuss the impact of the developed method.}, archive = {J_EJOR}, author = {Zeyu Liu and Xueping Li and Anahita Khojandi}, doi = {10.1016/j.ejor.2025.08.038}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {174-188}, shortjournal = {Eur. J. Oper. Res.}, title = {A primal–dual policy iteration algorithm for constrained markov decision processes}, volume = {328}, year = {2026}, }
• (2026). Controlling antithetic variates. EJOR, 328(1), 162-173. (www)

  @article{ , abstract = {We establish and investigate a theoretical framework for controlling covariance matrices in the method of antithetic variates through control variates to further reduce estimator variance. Instead of preemptively and carefully designing an estimator vector with negatively correlated components, the proposed framework starts with a predefined estimator vector that incorporates specified control variates. The weights and control matrix are then analytically determined through matrix algebra. The joint optimality of the resulting estimator variance is ensured with respect to both the weights and the control matrix, with closed-form implementable formulas derived for the optimal parameter pair. Numerical results are provided for various typical examples to illustrate the effectiveness, potential, and challenges of the proposed framework.}, archive = {J_EJOR}, author = {Reiichiro Kawai}, doi = {10.1016/j.ejor.2025.08.027}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {162-173}, shortjournal = {Eur. J. Oper. Res.}, title = {Controlling antithetic variates}, volume = {328}, year = {2026}, }
• (2026). Robust multi-period blood inventory routing under multiple uncertainties. EJOR, 328(1), 137-161. (www)

  @article{ , abstract = {We study a multi-period blood inventory routing problem that integrates production, inventory, and distribution decisions under uncertainties in demand, donation supply, and travel times, all while accounting for the limited shelf life of blood products. Our model captures transportation efficiency through a disutility measure based on vehicles’ arrival times at hospitals, and addresses supply–demand imbalances by allowing selective rejection of service requests at a high penalty cost. We formulate a robust optimization model that simultaneously determines production quantities, inventory levels, hospital service selections, and vehicle routing for each period. The objective is to minimize the total cost over the planning horizon, which includes worst-case inventory holding, wastage, and transportation costs, unserved demand penalties, and overall transportation disutility. To obtain an exact solution, we propose an integrated algorithm within the L -shaped framework that combines Benders decomposition with a branch-and-price-and-cut (BPC) scheme. This approach decomposes the robust model into a master problem and period-specific subproblems. For a given master solution, we first use constraint programming to verify the feasibility of the subproblems, and then, if feasible, solve them with a tailored BPC algorithm to generate Benders cuts that eliminate suboptimal master solutions. Extensive numerical experiments, including a case study at the Blood Center in Chongqing, demonstrate the effectiveness of our approach. Our analysis quantifies the benefits of incorporating uncertainty and robustness while providing managerial insights through a systematic evaluation of various parameters.}, archive = {J_EJOR}, author = {Ling Qing and Yunqiang Yin and Joshua Ignatius and Dujuan Wang}, doi = {10.1016/j.ejor.2025.05.036}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {137-161}, shortjournal = {Eur. J. Oper. Res.}, title = {Robust multi-period blood inventory routing under multiple uncertainties}, volume = {328}, year = {2026}, }
• (2026). Data-driven joint optimization of maintenance and spare parts provisioning: A distributionally robust approach. EJOR, 328(1), 122-136. (www)

  @article{ , abstract = {This paper investigates the joint optimization of condition-based maintenance and spare provisioning, incorporating insights obtained from sensor data. Prognostic models estimate components’ remaining lifetime distributions (RLDs), which are integrated into an optimization model to coordinate maintenance and spare provisioning. The existing literature addressing this problem assumes that prognostic models provide accurate estimates of RLDs, thereby allowing a direct adoption of Stochastic Programming or Markov Decision Process methodologies. Nevertheless, this assumption often does not hold in practice since the estimated distributions can be inaccurate due to noisy sensors or scarcity of training data. To tackle this issue, we develop a Distributionally Robust Chance Constrained (DRCC) formulation considering general discrepancy-based ambiguity sets that capture potential distribution perturbations of the estimated RLDs. The proposed formulation admits a Mixed-Integer Linear Programming (MILP) reformulation, where explicit formulas are provided to simplify the general discrepancy-based ambiguity sets. Finally, for the numerical illustration, we test a type- ∞ Wasserstein ambiguity set and derive closed-form expressions for the parameters of the MILP reformulation. The efficacy of our methodology is showcased in a wind turbine case study, where the proposed DRCC formulation outperforms other benchmarks based on stochastic programming and robust optimization.}, archive = {J_EJOR}, author = {Heraldo Rozas and Weijun Xie and Nagi Gebraeel and Stephen Robinson}, doi = {10.1016/j.ejor.2025.06.025}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {122-136}, shortjournal = {Eur. J. Oper. Res.}, title = {Data-driven joint optimization of maintenance and spare parts provisioning: A distributionally robust approach}, volume = {328}, year = {2026}, }
• (2026). Towards blockchain-enabled circular closed-loop supply chain and impact of consumers’ distrust in price, product greenness sensitivity and carbon tax and subsidy. EJOR, 328(1), 105-121. (www)

  @article{ , abstract = {With the increasing emphasis on environmental sustainability, both governments and consumers are more concerned than ever about the greenness of products. In this complex landscape, Supply Chains (SCs) face challenges in building trust and avoiding greenwashing accusations. Blockchain technology offers a promising solution by ensuring transparency and circularity within SCs, particularly in identifying customers for product recycling. This study pioneers the exploration of consumers' distrust in pricing and product greenness, alongside the impact of carbon policies (taxes and subsidies) within a closed-loop supply chain (CLSC). Using classical Stackelberg game theory, we develop two models that identify equilibrium decisions for SC members, focusing on pricing, green production investment, circularity, and blockchain adoption. Additionally, we propose an evolutionary game theory model to find the optimal government policies and identify the long-term behaviour of the CLSC and government in two heterogeneous populations. Our findings reveal that if the retailer's share of blockchain costs falls below a certain threshold, blockchain adoption becomes less profitable than exclusive investment in green production. A higher (lower) subsidy rate benefits (harms) the retailer but disadvantages (benefits) the collector. Blockchain adoption is generally more profitable for manufacturers and retailers, though less so for collectors, and it also drives greater investment in green production. While subsidies encourage blockchain adoption, they are not a sustainable long-term strategy for governments. Ultimately, the evolutionarily stable strategy for SCs involves a balanced investment in both green production and blockchain or green production alone, depending on market characteristics and cost-sharing structures.}, archive = {J_EJOR}, author = {Mohammad Akbarzadeh Sarabi and Ata Allah Taleizadeh and Arijit Bhattacharya}, doi = {10.1016/j.ejor.2025.06.030}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {105-121}, shortjournal = {Eur. J. Oper. Res.}, title = {Towards blockchain-enabled circular closed-loop supply chain and impact of consumers’ distrust in price, product greenness sensitivity and carbon tax and subsidy}, volume = {328}, year = {2026}, }
• (2026). A split-embedded metaheuristic for the heterogeneous inventory routing problem with batch size. EJOR, 328(1), 91-104. (www)

  @article{ , abstract = {Handling transportation and inventory management simultaneously is a challenging problem. The most famous optimization problem in this domain, known as the Inventory Routing Problem (IRP), aims to determine the routes and quantities to be delivered by a set of vehicles to meet customer demands at a minimum total inventory and transportation cost. The vast majority of works carried out so far on the IRP consider a homogeneous fleet of vehicles. This paper addresses, instead, a new IRP variant that considers intrinsic characteristics of real supply chains, such as a period-dependent heterogeneous fleet of vehicles and batch sizes for the delivered quantities. We model the problem with a Mixed Integer Linear Programming (MILP) formulation and propose a Split-Embedded Metaheuristic with a Post-Optimization phase (SEMPO) to solve it. Extensive computational experiments are conducted on a set of 80 new benchmark instances with up to 183 customers and a challenging time horizon of up 7 to 28 time periods to evaluate the performance of our approaches. The proposed SEMPO algorithm provides high-quality solutions and faster convergence compared to the MILP formulation.}, archive = {J_EJOR}, author = {Diego Perdigão Martino and Philippe Lacomme and Katyanne Farias}, doi = {10.1016/j.ejor.2025.05.044}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {91-104}, shortjournal = {Eur. J. Oper. Res.}, title = {A split-embedded metaheuristic for the heterogeneous inventory routing problem with batch size}, volume = {328}, year = {2026}, }
• (2026). The role of scarcity behavior in inventory management. EJOR, 328(1), 78-90. (www)

  @article{ , abstract = {A well-known phenomenon related to inventory, but often neglected in inventory management, is the scarcity effect, i.e., the increase in the demand if inventory is low. We consider a repeated purchase setting with a single firm and multiple buyers and address the question of whether inventory management decisions concerning the control policy and the service configuration (determining when and how much to order) impact scarcity behavior arising from buyers’ stock-out perception. We study common inventory control policies that assume a demand independent of the inventory management, and we challenge this critical assumption of an exogenous demand. This research explores two prevalent classes of inventory policies widely used in practice (periodic and continuous) configured according to fill rates, a popular way of measuring service. We conduct a laboratory experiment with four automated inventory management treatments (2 policies × 2 configurations) where participants act as buyers. We observe stock-out induced scarcity and find support for the hypothesis that the periodic policy leads to a stronger effect compared to the continuous policy if the service level is low. The study also supports the hypothesis that buyers act forward-looking as their demand peaks before the inventory reaches its lowest level. Overall, our research provides a new perspective on inventory management as it reveals that the chosen control policy and the selected service configuration influence stock-out pressure induced inventory runs. Inventory managers should be aware of scarcity effects and its consequences, like the disadvantages of a periodic policy for low service level.}, archive = {J_EJOR}, author = {Sebastian Schiffels and Christian Jost}, doi = {10.1016/j.ejor.2025.05.043}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {78-90}, shortjournal = {Eur. J. Oper. Res.}, title = {The role of scarcity behavior in inventory management}, volume = {328}, year = {2026}, }
• (2026). Green scheduling with time-of-use tariffs and machine states: Optimizing energy cost via branch-and-bound and bin packing strategies. EJOR, 328(1), 64-77. (www)

  @article{ , abstract = {This paper presents a branch-and-bound algorithm, enhanced with bin packing strategies, for scheduling under variable energy pricing and power-saving states. The proposed algorithm addresses the 1 , TOU | states | TEC problem, which involves scheduling jobs to minimize total energy cost (TEC) while considering time-of-use (TOU) electricity prices and different machine states (e.g., processing, idle, off). Key innovations include instance pre-processing for rapid lower bound calculations, a novel branching scheme combined with initializations, a block-finding primal heuristic, and a tighter lower bound for jobs with non-coprime processing times. These enhancements result in an efficient algorithm capable of solving benchmark instances with real energy prices with 200 jobs more than 100 times faster than existing state-of-the-art methods.}, archive = {J_EJOR}, author = {Ondřej Benedikt and István Módos and Antonin Novak and Zdeněk Hanzálek}, doi = {10.1016/j.ejor.2025.06.026}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {64-77}, shortjournal = {Eur. J. Oper. Res.}, title = {Green scheduling with time-of-use tariffs and machine states: Optimizing energy cost via branch-and-bound and bin packing strategies}, volume = {328}, year = {2026}, }
• (2026). Flight test scheduling: A generic model, lower bounds, and iterated local search. EJOR, 328(1), 49-63. (www)

  @article{ , abstract = {Flight tests play a critical role in the R&D process for new aircraft as they help verify the airworthiness and capabilities and expose design and manufacturing defects. During the test course, a large number of tasks need to be scheduled appropriately so that the flight tests can be completed with minimum cost and high efficiency. Therefore, there is a strong need for developing an efficient method that can generate high-quality test schedules. In this paper, we study flight test scheduling to minimize the number of required test flights, thereby decreasing the cost and time required during the entire test course. We establish a mixed-integer programming model to formally describe the problem, propose several computationally efficient lower bounds to help verify the quality of obtained solutions, and develop an iterated local search algorithm for generating a high-quality solution in an effective manner. Comprehensive computational experiments are performed to demonstrate the efficiency of our proposed algorithm. We report some general managerial insights based on the obtained computational results.}, archive = {J_EJOR}, author = {Hanqiao Tao and Guopeng Song and Roel Leus and Zhe Liang and Jiang Jiang}, doi = {10.1016/j.ejor.2025.06.001}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {49-63}, shortjournal = {Eur. J. Oper. Res.}, title = {Flight test scheduling: A generic model, lower bounds, and iterated local search}, volume = {328}, year = {2026}, }
• (2026). Nested logic-based benders decomposition for an integrated home healthcare problem. EJOR, 328(1), 32-48. (www)

  @article{ , abstract = {In this study, we apply nested logic-based Benders decomposition to solve an integrated home healthcare staffing, assignment, routing, and scheduling problem with application in Norway. The proposed method operates at two decomposition levels. Consequently, the entire problem is decomposed into three hierarchical sub-problems: the staffing problem, the assignment problem, and the routing and scheduling problem. These sub-problems are interrelated through two levels of logic-based Benders cuts. Computational experiments on 40 test instances demonstrate the superior performance of nested logic-based Benders decomposition compared to directly solving a mixed-integer linear programming model available in the literature. Specifically, the proposed solution method achieved proven optimality in 28 instances and provided feasible solutions for the remaining 12 instances. In contrast, directly solving the mixed-integer linear programming model yielded proven optimality in 16 instances, provided feasible solutions for 20 instances, and failed to find feasible solutions for 4 instances within the same computational time limit.}, archive = {J_EJOR}, author = {Abdalrahman Algendi and Sebastián Urrutia and Lars Magnus Hvattum and Rafael A. Melo}, doi = {10.1016/j.ejor.2025.06.006}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {32-48}, shortjournal = {Eur. J. Oper. Res.}, title = {Nested logic-based benders decomposition for an integrated home healthcare problem}, volume = {328}, year = {2026}, }
• (2026). A reduction approach for the parallel machine scheduling problem with a separate server for loading and unloading operations. EJOR, 328(1), 15-31. (www)

  @article{ , abstract = {This paper addresses the non-preemptive identical parallel machine scheduling problem with a dedicated loading server and a dedicated unloading server. Each job has to be loaded by the dedicated loading server immediately before being processed on one of the identical parallel machines, and unloaded immediately by the dedicated unloading server after its processing. The objective function involves the minimization of the makespan. This problem arises in the semiconductor industry, plastic injection industry, kitchen production systems, healthcare, container terminals, and many other industrial fields. We prove the problem to be strongly NP-hard, analyze a special case with identical loading, processing, and unloading times, and establish a tight lower bound. In addition, we propose two novel mixed-integer linear programming formulations: one utilizing time-indexed variables with an iterative strengthening algorithm, and the other employing linear-ordering variables along with two enhanced valid inequalities. Given the complexity of the problem, we introduce a reduction approach that simplifies the problem by modifying certain constraints, enabling the determination of a feasible solution much more quickly. Building on this reduction, we provide a linear-time reduction algorithm and a fast formulation based on assignment-and-positional date variables. Furthermore, we study a special case involving regular jobs. To solve large-scale instances of the problem with up to 250 jobs and up to 5 machines, we design a hybrid approach combining an iterated greedy algorithm with variable neighborhood descent. As shown in the computational experiments on two sets of benchmark instances, the reduction approach significantly outperforms all previous methods existing in the literature.}, archive = {J_EJOR}, author = {Abdelhak Elidrissi and Jatinder N.D. Gupta and Rachid Benmansour and Bertrand M.T. Lin}, doi = {10.1016/j.ejor.2025.05.030}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {15-31}, shortjournal = {Eur. J. Oper. Res.}, title = {A reduction approach for the parallel machine scheduling problem with a separate server for loading and unloading operations}, volume = {328}, year = {2026}, }
• (2026). Cutting stock problem with usable leftovers: A review. EJOR, 328(1), 1-14. (www)

  @article{ , abstract = {This article presents a comprehensive literature review of the Cutting Stock Problem with Usable Leftovers (CSPUL). The most recent review on this topic dates to 2014, covering articles published before 2013. Since then, the number of publications on CSPUL has increased significantly, driven by new applications and more efficient solution approaches. We analyze fifty two relevant articles from twenty four different journals, focusing on works published after 2008 while acknowledging foundational contributions from the 1980s and 1990s. This review categorizes variations of CSPUL based on their dimensions (1D, 2D, and 3D), planning period characteristics (single-period and multi-period), objective functions, and solution methods. The article provides a detailed summary of the key features in the mathematical models and solution methods proposed in these studies. Additionally, it highlights several industrial applications of CSPUL, illustrating its practical relevance. Through this analysis, we identify important applications and propose promising directions for future research. The findings and insights presented here have practical implications for optimizing resource utilization and promoting sustainability in industries facing cutting challenges.}, archive = {J_EJOR}, author = {Victor Senergues and Nadjib Brahimi and Adriana Cristina Cherri and François Klein and Olivier Péton}, doi = {10.1016/j.ejor.2025.03.014}, journal = {European Journal of Operational Research}, month = {1}, number = {1}, pages = {1-14}, shortjournal = {Eur. J. Oper. Res.}, title = {Cutting stock problem with usable leftovers: A review}, volume = {328}, year = {2026}, }

ESWA - 212

• (2026). RAFN: A risk-aware feature network for identifying risk factors in supply chain finance. ESWA, 298, 129874. (www)

  @article{ , abstract = {As supply chain finance businesses expand, traditional risk assessment systems, which rely heavily on manual processes and static rule-based frameworks, are increasingly unable to keep up with the complexity and dynamism of modern risk patterns. This often leads to delayed responses and inefficiencies in risk management. To address key challenges such as difficulties in integrating heterogeneous data, low detection rates for hidden risks, and limited ability to capture dynamic risk patterns, this paper introduces a novel Risk-Aware Feature Network (RAFN) driven by an adaptive attention mechanism. The RAFN model is designed with a dual-channel architecture to process numerical and categorical data separately, employs gated linear units to dynamically merge heterogeneous data streams, and incorporates a multi-head attention mechanism with dynamic coefficients to focus on risk-sensitive features adaptively. Experiments conducted on both public and proprietary datasets show that RAFN outperforms mainstream algorithms, achieving a 1.73%-5.81% improvement in accuracy, recall, and F1-score, while maintaining a strong balance between specificity and recall. Furthermore, this study proposes a closed-loop risk management framework based on RAFN, which integrates “smart contract triggering, off-chain model evaluation, and on-chain consensus validation.” This approach offers an efficient technical solution to break down data silos and enhance the precision of risk identification in supply chain finance, paving the way for more effective and reliable risk control systems.}, archive = {J_ESWA}, author = {Yang Zhang and Yating Zhao and Wenjuan Lian and Bin Jia}, doi = {10.1016/j.eswa.2025.129874}, journal = {Expert Systems with Applications}, month = {3}, pages = {129874}, shortjournal = {Expert Syst. Appl.}, title = {RAFN: A risk-aware feature network for identifying risk factors in supply chain finance}, volume = {298}, year = {2026}, }
• (2026). Photovoltaic cluster power ultra-short-term cross-seasonal prediction integrating multi-channel information probabilistic diffusion generation and improved offset loss strategy. ESWA, 298, 129826. (www)

  @article{ , abstract = {Accurate photovoltaic (PV) power prediction under complex meteorological conditions remains challenging, particularly given the pronounced seasonal variations that obscure generation patterns. This study presents a novel ultra-short-term prediction framework integrating meteorological volatility analysis with seasonal characteristic modeling. We developed a specialized multi-channel Gram angular summation field (MGASF) transformation matrix to holistically capture meteorological fluctuations, subsequently leveraging denoising diffusion probabilistic model (DDPM) for strategic augmentation of under-represented weather scenarios to enhance similar-day identification. Our hybrid architecture combines multi-channel vision Transformer (VIT) with bidirectional long and short-term memory (BILSTM) networks to synergistically analyze temporal dependencies and spatial patterns in PV similarity recognition. Furthermore, we engineered a seasonal-adaptive prediction system through an improved variable-weight Smooth L1 loss function, establishing an optimized seasonal alignment mechanism that achieves high-precision prediction across varying meteorological conditions with minimal computational overhead. Through rigorous validation using operational data from a utility-scale photovoltaic cluster in Western Inner Mongolia, the proposed method achieved consistent accuracy improvements: 3.02 % reduction in N RMSE , 1.65 % decrease in N MAE , and 2.19 % enhancement in R 2 compared to baseline approaches in PV cluster. These statistically significant enhancements demonstrate our framework’s capability to mitigate seasonal impacts while maintaining prediction reliability in complex meteorological environments.}, archive = {J_ESWA}, author = {Mao Yang and Yue Jiang and Yunfeng Guo and Jianfeng Che and Wei He and Kang Wu}, doi = {10.1016/j.eswa.2025.129826}, journal = {Expert Systems with Applications}, month = {3}, pages = {129826}, shortjournal = {Expert Syst. Appl.}, title = {Photovoltaic cluster power ultra-short-term cross-seasonal prediction integrating multi-channel information probabilistic diffusion generation and improved offset loss strategy}, volume = {298}, year = {2026}, }
• (2026). Sustainable time-dependent intermodal hub-and-spoke logistic network considering hub failure: A mathematical model and a hybrid artificial bee colony algorithm. ESWA, 298, 129804. (www)

  @article{ , abstract = {This paper addresses the design of a sustainable hub-and-spoke logistics network that integrates intermodal transportation between the hubs, hub failures, time dependency, and environmental parameters. Accordingly, we propose a novel mixed-integer linear programming (MILP) model and a hybrid artificial bee colony-based algorithm (HABCb) to minimize transportation costs and emissions in robust network configurations. The model is the first to simultaneously integrate intermodality, sustainability metrics, and hub disruption scenarios within a single framework. Computational experiments using real-life data from Turkey demonstrate that the proposed HABCb approach outperforms both genetic algorithm (GA) and artificial bee colony (ABC) algorithm. On medium-sized problem sets, it achieves average cost reductions of 7% compared to GA and 10% compared to ABC algorithm, while on large-sized problems the reductions are 10% and 15%, respectively. Furthermore, the HABCb approach provides faster convergence and higher-quality solutions for larger problem sizes. The findings highlight the practical and theoretical insights of incorporating sustainability, intermodality, and robustness into hub-and-spoke network design.}, archive = {J_ESWA}, author = {Burcu Tokbay Erkek and Salih Himmetoğlu and Yılmaz Delice and Emel Kızılkaya Aydoğan}, doi = {10.1016/j.eswa.2025.129804}, journal = {Expert Systems with Applications}, month = {3}, pages = {129804}, shortjournal = {Expert Syst. Appl.}, title = {Sustainable time-dependent intermodal hub-and-spoke logistic network considering hub failure: A mathematical model and a hybrid artificial bee colony algorithm}, volume = {298}, year = {2026}, }
• (2026). Adaptive gated universal information extraction for chinese legal texts. ESWA, 298, 129801. (www)

  @article{ , abstract = {Constructing knowledge graphs in legal domains requires simultaneous extraction of entities and relations. To reduce repeated modeling in traditional approaches, we adopt the Universal Information Extraction (UIE) model as a foundation and propose an enhanced variant named Adaptive Gated Universal Information Extraction (AGUIE). This study develops a new decoder based on the Adaptive Focusing Gated Attention Unit (AFGAU). This unit enhances the standard Gated Attention Unit (GAU) by integrating two key components—learnable dynamic convolution and reset/update gating mechanisms. Moreover, the study employs a cross-pointer structure as the output layer to better identify information boundaries. To support this study, we construct a domain specific dataset for extracting key information from legal judgment documents. Systematic comparative analysis and ablation studies demonstrate that AGUIE achieves significant performance gains over baseline UIE, with an F1 score of 85.56% on our legal judgment documents dataset. Additionally, we evaluate the model’s generalization on public datasets such as ACE04, ACE05, and CoNLL04, covering both entity recognition and relation extraction tasks. Experimental results indicate that AGUIE demonstrates competitive results with recent studies on ACE04-Ent and CoNLL04, outperforms them on the ACE05 dataset, achieving F1 scores of 87.19% on ACE05-Ent and 79.29% on ACE05-Rel. In conclusion, AGUIE is a reliable and effective solution for universal information extraction in both legal and general domains.}, archive = {J_ESWA}, author = {Yabo Liu and Yatong Zhou and Kuo-Ping Lin}, doi = {10.1016/j.eswa.2025.129801}, journal = {Expert Systems with Applications}, month = {3}, pages = {129801}, shortjournal = {Expert Syst. Appl.}, title = {Adaptive gated universal information extraction for chinese legal texts}, volume = {298}, year = {2026}, }
• (2026). Spatiotemporal online fuzzy modeling with knowledge-driven differential evolution automatic clustering for distributed parameter systems. ESWA, 298, 129785. (www)

  @article{ , abstract = {Distributed parameter systems are prevalent in various industrial processes and attract significant attention. However, these systems exhibit complex spatiotemporal coupling characteristics, and effectively determining the fuzzy rules of the antecedent set is crucial for improving modeling performance. Traditional clustering methods typically rely on empirical heuristics and are unable to adapt to dynamic system characteristics under changing environments. In high-dimensional and nonlinear scenarios, the number of fuzzy rule combinations grows exponentially, significantly increasing computational complexity. Therefore, an online spatiotemporal three-dimensional fuzzy modeling method based on knowledge-driven differential evolution automatic clustering and extreme learning machine (3D-OSADE-ELM) is proposed for the complex nonlinear distributed parameter system. First, an automatic clustering mechanism based on differential evolution and extreme learning machine initializes the fuzzy rules within the three-dimensional fuzzy system. Subsequently, a knowledge-driven archiving mechanism dynamically updates the fuzzy rules of the antecedent set during the online incremental learning phase. Finally, the spatial basis function is obtained by learning the output weight of the online extreme learning machine. The validation experiments conducted on the rapid thermal chemical vapor deposition reactor system and the nonisothermal packed-bed system demonstrate the effectiveness and superiority of the proposed method.}, archive = {J_ESWA}, author = {Gang Zhou and Xianxia Zhang and Bing Wang}, doi = {10.1016/j.eswa.2025.129785}, journal = {Expert Systems with Applications}, month = {3}, pages = {129785}, shortjournal = {Expert Syst. Appl.}, title = {Spatiotemporal online fuzzy modeling with knowledge-driven differential evolution automatic clustering for distributed parameter systems}, volume = {298}, year = {2026}, }
• (2026). Time-balanced MSE for machinery imbalanced degradation trend prediction. ESWA, 298, 129783. (www)

  @article{ , abstract = {Accurate degradation trend prediction (DTP) is crucial for optimizing equipment operation and maintenance. With the rapid development of artificial intelligence, many data-driven methods have been applied to machinery degradation trend prediction. In practice, most machineries are in the early stages of degradation, with only a few reaching the final stages, leading to a temporal imbalanced data distribution. Current research on imbalanced distributions mainly focuses on classification tasks. However, DTP involves multiple time-dependent continuous targets, making classification-based methods unsuitable. To address this issue, the degradation trend prediction task is reformulated as a multi-task problem and a novel time-balanced Mean Square Error (TBMSE) loss function is proposed. In each prediction task, the Gaussian Mixture Model (GMM) is used to fit the training label distribution. Additionally, the cumulative information noise for each prediction task is modeled using GMM, and an end-to-end network structure is designed to learn the GMM parameters. Experiments are conducted on the IMS bearing dataset and the turboprop engine dataset, demonstrating that the TBMSE loss effectively mitigates the issue of temporal imbalanced distribution in degradation trend prediction.}, archive = {J_ESWA}, author = {Yu-Qiang Wang and Yong-Ping Zhao and Tian-Ding Zhang and Yu-Wei Wang}, doi = {10.1016/j.eswa.2025.129783}, journal = {Expert Systems with Applications}, month = {3}, pages = {129783}, shortjournal = {Expert Syst. Appl.}, title = {Time-balanced MSE for machinery imbalanced degradation trend prediction}, volume = {298}, year = {2026}, }
• (2026). RSUTrajRec: Multi-granularity trajectory recovery based on roadside units sensing. ESWA, 298, 129780. (www)

  @article{ , abstract = {Vehicle mobility trajectories, especially fine-grained trajectories, provide valuable insights for understanding urban dynamics and play a crucial role in intelligent transportation systems and urban planning. Obtaining fine-grained vehicle trajectories can be realized by trajectory recovery, but traditional efforts suffer from defects such as poor privacy protection and low recovery accuracy. To address these issues, we propose a new scenario of trajectory recovery based on roadside unit (RSU) sensing. However, this scenario introduces a significant challenge: recovering high-precision trajectories from the incomplete and unevenly distributed sensing data. To tackle this, we design RSUTrajRec , a multi-granularity trajectory recovery framework that comprises a graph neural network-based module for road information prediction, a Transformer-based module for multi-granularity recovery, and an RSU deployment planning module. Extensive real-world dataset evaluations reveal that RSUTrajRec has a significant advantage in recovering missing vehicle trajectories outside the RSU coverage area. In addition, evaluations also verify that the performance of the trajectory recovery task can be effectively improved by optimizing the RSU deployment plan.}, archive = {J_ESWA}, author = {Xianjing Wu and Xutao Chu and Jianyu Wang and Shengjie Zhao}, doi = {10.1016/j.eswa.2025.129780}, journal = {Expert Systems with Applications}, month = {3}, pages = {129780}, shortjournal = {Expert Syst. Appl.}, title = {RSUTrajRec: Multi-granularity trajectory recovery based on roadside units sensing}, volume = {298}, year = {2026}, }
• (2026). Technology foresight in china’s industrial robotics with MLWS-TF: A machine learning and weak signal-based system. ESWA, 298, 129779. (www)

  @article{ , abstract = {Technology foresight analyses technological trends and potential impacts to provide strategic guidance. However, existing methods either rely on experts to discover emerging directions leading to subjective bias or adopt machine learning to predict without explanation. We propose a Machine Learning and Weak Signal-based Technology Forecasting System (MLWS-TF), which is entirely data-driven to enhance the objectivity of technology foresight and can interpret emerging directions through weak signals. The system adopts a two-phase machine learning model (2P-ML), the first phase identifies papers related to the robotics field, while the second further classifies them into fine-grained research directions. Keywords are extracted from the papers using a Word2Vec-based approach, and a three-dimensional signal classification method (DVI) is developed to quantify the foresight value of keywords across the Diffusion, Visibility, and Impact dimensions, identifying weak signals for technology forecasting. Experiments evaluate various machine learning algorithms, and XGBoost outperforms in constructing the 2P-ML classifier. The model achieved over 90% accuracy, demonstrating its effectiveness in identifying the theme of scientific documents based on textual features. For each research theme, the DVI provides a more comprehensive assessment of signal strength to detect weak signals. Finally, MLWS-TF analyses the growth potential of themes and successfully identifies critical development directions. Our approach offers a novel automated technology foresight system, which completely avoids the subjectivity and dependence on expert judgment that characterize traditional technology foresight approaches, and extends weak signal theory by introducing the Impact dimension to evaluate signal strength.}, archive = {J_ESWA}, author = {Ruihan Wang and Yuhao Zhu}, doi = {10.1016/j.eswa.2025.129779}, journal = {Expert Systems with Applications}, month = {3}, pages = {129779}, shortjournal = {Expert Syst. Appl.}, title = {Technology foresight in china’s industrial robotics with MLWS-TF: A machine learning and weak signal-based system}, volume = {298}, year = {2026}, }
• (2026). Generating realistic pruning solutions for automated grape vine pruning using graph neural networks. ESWA, 298, 129778. (www)

  @article{ , abstract = {In our prior work we showed that graph neural networks (GNNs) can be trained to generate pruning solutions that could direct robotic pruning robots to perform automated cane pruning of wine grape vines. That study introduced the feasibility of the technology but also showed that there were many open questions and issues with the research results that needed to be addressed. In this study we address some of these questions. For example, we answer the question of how would a model like this perform on real vine architectures compared with pruning solutions from real experienced pruners. Our most notable contributions include moving away from a per-cane classification model that attempts to define a single perfect pruning solution, to a model that ranks multiple good solutions and picks the best one. We addressed a key limitation of the previous training data by moving away from synthetic vine architectures to realistic ones recorded from real vines and using pruning solutions collected by expert pruners as our ground-truth. Our primary goal was to show that learning by example using a GNN-based model was a viable approach to automated pruning, even when compared with experienced pruners. We showed robust performance from our model by training on a dataset of 90 pruning solutions generated by expert pruners in the 2022 season, and testing our performance on 117 pruning solutions from an independent set of pruners from the 2021 season. The model was able to correctly score all the pruning solutions from the 2021 dataset as good to very good and none of the expert solutions were classified as poor .}, archive = {J_ESWA}, author = {Jaco Fourie and Jeffrey Hsiao and Oliver Batchelor and Kevin Langbroek and Henry Williams and Richard Green and Armin Werner}, doi = {10.1016/j.eswa.2025.129778}, journal = {Expert Systems with Applications}, month = {3}, pages = {129778}, shortjournal = {Expert Syst. Appl.}, title = {Generating realistic pruning solutions for automated grape vine pruning using graph neural networks}, volume = {298}, year = {2026}, }
• (2026). Mining spatiotemporal dominant co-location patterns. ESWA, 298, 129775. (www)

  @article{ , abstract = {Spatial co-location pattern mining is an important branch of spatial data mining, which can identify spatial features that prevalently occur in proximity. Based on spatial co-location patterns, the research of dominant relationships mining within co-location patterns further considers the influence relationship among features. However, relying solely on spatial data to analyze the positions and distribution of features for mining dominant relationships is insufficient and may lead to incorrect patterns. To address this limitation, this paper introduces the temporal factor into the research of dominant relationships mining and proposes the spatiotemporal dominant co-location pattern mining (STDCPM). At first, we define the concepts of spatiotemporal dominant relationship from both temporal and spatial dimensions, and then propose the spatiotemporal dominant participation index to assess the prevalence of spatiotemporal dominant co-location patterns. Furthermore, we design two algorithms, the spatiotemporal dominant co-location pattern mining algorithm with level-by-level search and its improved version, i.e., the spatiotemporal dominant co-location pattern mining approach based on dual pruning and refining set (STDCPM-DPR), to ensure efficient mining in spatiotemporal datasets. The time complexity, correctness, and completeness of proposed algorithms are discussed. Extensive experiments on real-world datasets demonstrate the effectiveness of STDCPM and the efficiency of STDCPM-DPR algorithm.}, archive = {J_ESWA}, author = {Jiangchuan Mei and Peizhong Yang and Hongmei Chen and Lizhen Wang}, doi = {10.1016/j.eswa.2025.129775}, journal = {Expert Systems with Applications}, month = {3}, pages = {129775}, shortjournal = {Expert Syst. Appl.}, title = {Mining spatiotemporal dominant co-location patterns}, volume = {298}, year = {2026}, }
• (2026). Cross-scale content adaptive network for three-dimensional multi-object tracking and fish activity quantification. ESWA, 298, 129774. (www)

  @article{ , abstract = {Tracking and quantifying fish activity are vital for evaluating their health status and adaptability to the environment. However, most current research on fish tracking and activity quantification suffers from the limitation of being two-dimensional, losing crucial vertical or horizontal information. To facilitate tracking and quantitative analysis of fish activity in three-dimensional (3D) space, a cross-scale content-adaptive network-based 3D multi-object tracking method for fish is proposed, through which fish movements are quantified accordingly. Firstly, a cross-scale content-adaptive fusion network is proposed to accurately determine the fish positions from top-down and side views, thereby mitigating the issue of scale variation across different perspectives. Secondly, a hierarchical tracking method is implemented to obtain the 3D trajectories of the fish, addressing the challenge of cross-view identity matching. Finally, activity parameters in 3D space, including the activity quantity and trajectory length for individual fish, as well as the dispersion and cohesion for the fish group, are calculated. The proposed method was validated, achieving a Multi-Object Tracking Accuracy (MOTA) of 97.68% and an Identification F1 Score (IDF1) of 97.93%. For activity quantification, the Mean Absolute Error (MAE) was found to be 0.088 (unit weight·(cm/s) 2 ), and the Root Mean Square Error (RMSE) was 0.1064 (unit weight·(cm/s) 2 ). These results affirm the method’s adaption of fish features across scales for 3D tracking and activity analysis. With its efficient performance, our method presents as an instrument for activities such as fish behavior monitoring, selective breeding, and environmental assessment.}, archive = {J_ESWA}, author = {Yiran Liu and Dingshuo Liu and Mingrui Kong and Beibei Li and Qingling Duan}, doi = {10.1016/j.eswa.2025.129774}, journal = {Expert Systems with Applications}, month = {3}, pages = {129774}, shortjournal = {Expert Syst. Appl.}, title = {Cross-scale content adaptive network for three-dimensional multi-object tracking and fish activity quantification}, volume = {298}, year = {2026}, }
• (2026). A novel fuzzy clustering approach with transition matrix for explainable evaluation of social media-based digital literacy interventions. ESWA, 298, 129769. (www)

  @article{ , abstract = {Assessing the effectiveness of digital literacy interventions often relies on raw score comparisons or hard classifications, which may obscure nuanced changes in conceptual understanding and provide limited interpretability. Traditional approaches fail to capture the probabilistic and fuzzy nature of learning progression and do not support transparent analysis of how learners transition across conceptual clusters over time. This study proposes an explainable evaluation framework that integrates fuzzy clustering with a fuzzy transition matrix to model the redistribution of aggregated membership values between pretest and posttest conceptual clusters. The framework applies Fuzzy C-Means (FCM) to derive soft cluster memberships and constructs a transition matrix that represents probabilistic learning progression in a linguistically interpretable form. Unlike conventional methods, this approach enables the analysis of gradual transitions across levels of proficiency rather than binary outcomes. The model was applied to real-world educational data from control and experimental classes, the latter of which received a social media-based instructional intervention. Results indicate that the control class exhibited downward or stagnant patterns, particularly among high-performing learners, while the experimental class showed more coherent upward cluster transitions among low- and moderate-level learners. By enabling interpretable modeling of pre–post cluster transition patterns, the proposed framework contributes to the advancement of explainable machine learning in education. It also highlights the potential of social computing platforms to foster scalable, data-driven digital literacy development.}, archive = {J_ESWA}, author = {Rustam and Diana Noor Anggraini and Koredianto Usman and Loveleen Gaur}, doi = {10.1016/j.eswa.2025.129769}, journal = {Expert Systems with Applications}, month = {3}, pages = {129769}, shortjournal = {Expert Syst. Appl.}, title = {A novel fuzzy clustering approach with transition matrix for explainable evaluation of social media-based digital literacy interventions}, volume = {298}, year = {2026}, }
• (2026). Dynamic graph structure correction with nonadjacent correlations for multivariate time series forecasting. ESWA, 298, 129768. (www)

  @article{ , abstract = {Effectively modeling the relations between variables in multivariate time series is of utmost importance for accomplishing accurate predictions. In real-world scenarios, in addition to sequential correlations, the evolution of relations between variables also exhibits nonadjacent correlations at different scales. However, existing methods primarily focus on constructing dynamic graph structures at each time step using temporal features extracted by continuous temporal models, which cannot capture above latent dependencies. In this study, we introduce the Dynamic Graph Structure Correction (DGC) model, leveraging a multi-scale framework with dilated convolution. To take full advantage of nonadjacent correlations in the evolution of relations between variables, we adaptively select history-related graph structures to correct initial graph structure constructed by Gate Recurrent Units. In addition, we design a time-decay-based attention mechanism to address the influence of time intervals between history-related and current time steps. Finally, the evolved graph structures are fed into graph neural networks to handle the multi-scale and complex structural relations. Our proposed model achieves superior performance compared to state-of-the-art methods in multivariate time series forecasting, as evidenced by the evaluation results on four widely used benchmark datasets.}, archive = {J_ESWA}, author = {Dandan He and Yueyang Wang and Chaoli Lou and Gang Tan and Qingyu Xiong and Guodong Sa}, doi = {10.1016/j.eswa.2025.129768}, journal = {Expert Systems with Applications}, month = {3}, pages = {129768}, shortjournal = {Expert Syst. Appl.}, title = {Dynamic graph structure correction with nonadjacent correlations for multivariate time series forecasting}, volume = {298}, year = {2026}, }
• (2026). Optimized homomorphic linear computation in privacy-preserving CNN inference. ESWA, 298, 129767. (www)

  @article{ , abstract = {Machine Learning as a Service (MLaaS) provides robust solutions for deploying deep learning inference in cloud environments. However, it also raises serious privacy concerns regarding user data and proprietary model parameters. Numerous hybrid cryptographic protocols that integrate homomorphic encryption (HE) and garbled circuits (GC) have been proposed to enable secure inference with low latency. In these protocols, the homomorphic evaluation of linear operations remains the primary performance bottleneck and warrants further optimization. In this work, we propose novel optimizations for HE-based linear computations within the hybrid cryptographic framework for secure neural network inference. Specifically, we devise two efficient strategies for homomorphic matrix-vector multiplication and convolution. For matrix-vector multiplication, we introduce a grouped diagonal extraction technique that encodes the weight matrix more compactly and enables configurable ciphertext rotation reuse, while for homomorphic convolution, we present a group-wise combine-and-merge evaluation method. Both methods significantly reduce the number of required ciphertext rotations. Our approach achieves up to a 3.9 × speedup in matrix-vector multiplication and a 2.9 × improvement in convolution over state-of-the-art (SOTA) solutions. The HE-GC hybrid secure convolutional neural networks (CNN) inference framework incorporating these enhancements yields speedups of 2.5 × on widely used ResNets deep learning architectures.}, archive = {J_ESWA}, author = {Chenglong Li and Xirong Ma and Xiuhao Wang and Fanyu Kong and Yunting Tao and Chunpeng Ge}, doi = {10.1016/j.eswa.2025.129767}, journal = {Expert Systems with Applications}, month = {3}, pages = {129767}, shortjournal = {Expert Syst. Appl.}, title = {Optimized homomorphic linear computation in privacy-preserving CNN inference}, volume = {298}, year = {2026}, }
• (2026). SS-RDHEI for embedding capacity enhancement by PDPM and auxiliary data free coding. ESWA, 298, 129766. (www)

  @article{ , abstract = {Reversible data hiding in encrypted images (RDHEI) is a promising technique for multimedia cloud computing that enables the embedding of secret data into encrypted images while preserving confidentiality. However, the existing RDHEI algorithms fail to meet the high-security requirements of distributed storage systems in the cloud. Although, secret sharing based RDHEI (SS-RDHEI) may solve this problem, the current methods have weakness such as insufficient embedding capacity and unsatisfactory balance between image security and redundancy. To enhance the algorithm’s ability to carry information, this paper proposes a SS-RDHEI for embedding capacity enhancement by PDPM and auxiliary data free coding. Firstly, pixel-difference preservation based modulation (PDPM) ensures secure encryption by modifying all pixels except for a reference block, minimizing damage; moreover, an improved block-level pixel predictor enhances carrier redundancy. Secondly, auxiliary data free coding (ADFC) marks prediction errors directly in the binary sequence of the original pixel without auxiliary information while maintaining accuracy, and reduces the impact of different textures on embedding performance byselecting optimal parameters for each share image. Finally, by combining PDPM with secret sharing, it achieves independent embedding for multiple data hiders while ensuring fair information embedding. Experimental results demonstrate that the proposed algorithm outperforms existing state-of-the-art schemes in terms of information-carrying capability.}, archive = {J_ESWA}, author = {Zhihua Gan and Zongwei Tang and Yalin Song and Gongyao Cao and Xiuli Chai and Yushu Zhang}, doi = {10.1016/j.eswa.2025.129766}, journal = {Expert Systems with Applications}, month = {3}, pages = {129766}, shortjournal = {Expert Syst. Appl.}, title = {SS-RDHEI for embedding capacity enhancement by PDPM and auxiliary data free coding}, volume = {298}, year = {2026}, }
• (2026). Improved convolutional neural networks for the bullwhip effect in supply chains. ESWA, 298, 129764. (www)

  @article{ , abstract = {The Bullwhip Effect (BWE) introduces significant challenges to production systems by amplifying demand and order oscillations. One of the most effective methods for predicting and modeling complex systems is Convolutional Neural Networks (CNNs). However, certain phenomena, such as the BWE in supply chains (SC), are difficult to predict and identify directly. The primary challenge for Machine Learning (ML) algorithms in this context lies in the training phase: the raw demand and order data are fed into the network, yet the desired training outcome is the oscillatory behavior of these data from the perspective of the BWE. Consequently, conventional max pooling, average pooling operators, kernels, and weighted linear combinations of data are insufficient for capturing this type of learning. To address this issue, in this paper, a novel structure containing new pooling operators and kernels of CNNs is proposed to tailor the unique characteristics of the BWE. Specifically: a ) Considering the temporal propagation nature of the BWE, new filters and pooling operators were designed to enable CNNs to predict the BWE accurately. b ) A tensor structure was also proposed for the time signal of demand as inputs of the CNNs to facilitate the analysis of all factors influencing the occurrence of the BWE. c ) To capture the magnitude of the BWE among features, a novel combination of filters and pooling operators was proposed, enabling the CNNs to account for hidden but yet significant feature effects during training. The benefits of the proposed approach lie in its versatility, and it can be applied to train CNNs to model structured fluctuations like the BWE in various dynamic systems.}, archive = {J_ESWA}, author = {Sajjad Aslani Khiavi and Farzad Hashemzadeh}, doi = {10.1016/j.eswa.2025.129764}, journal = {Expert Systems with Applications}, month = {3}, pages = {129764}, shortjournal = {Expert Syst. Appl.}, title = {Improved convolutional neural networks for the bullwhip effect in supply chains}, volume = {298}, year = {2026}, }
• (2026). A hybrid risk assessment method combining CatBoost and FAHP-grid search optimized risk matrix for container ship accident. ESWA, 298, 129763. (www)

  @article{ , abstract = {As a dominant mode of maritime transportation with unique risk characteristics, container shipping requires accurate and applicable risk assessment. However, conventional risk matrices oversimplify complex interactions, while pure data-driven models lack operational utility. To address this, a hybrid method for container ship risk assessment is proposed. This method integrates CatBoost-based predictive method, FAHP-grid search optimized risk matrix, and GIS-supported risk mapping. A comprehensive study of maritime casualties and piracy accidents is conducted, utilizing historical incident data sets collected from the Global Integrated Shipping Information System (GISIS). The global maritime accident risk of container ships is then evaluated and mapped. The sensitivity analysis confirms the robustness of the method under varying linguistic distance parameters, while expert weights have a moderate impact on the assessment results. Finally, the effectiveness of the proposed method is validated through comparative analyses on predictive performance, risk discrimination capability, and risk assessment accuracy. CatBoost algorithm outperforms XGBoost, LightGBM, and Random Forest algorithms in predictive metrics. The designed risk matrix shows strong discriminatory ability for container ship risk levels. In historical accident data validation, the proposed method also achieves higher accuracy than combinations involving XGBoost, LightGBM, or Random Forest with the designed risk matrix.}, archive = {J_ESWA}, author = {Yuqing Xiao and Shilian Han and Xinwang Liu}, doi = {10.1016/j.eswa.2025.129763}, journal = {Expert Systems with Applications}, month = {3}, pages = {129763}, shortjournal = {Expert Syst. Appl.}, title = {A hybrid risk assessment method combining CatBoost and FAHP-grid search optimized risk matrix for container ship accident}, volume = {298}, year = {2026}, }
• (2026). Quadrotor navigation considering attitude: A deep reinforcement learning method using tangent path rewards. ESWA, 298, 129762. (www)

  @article{ , abstract = {Autonomous navigation of quadrotors is a fundamental prerequisite for numerous applications. This work proposes a novel deep reinforcement learning (DRL) framework that explicitly addresses quadrotor attitude dynamics during autonomous navigation, a critical yet underexplored challenge in existing learning-based UAV navigation studies. In the proposed method, high-level velocity commands will be generated by a deep neural network policy and translated by a low-level control algorithm to achieve precise control of both positions and rotations of quadrotors. A specialized network structure is designed to effectively extract environmental obstacle features and quadrotor sequence features to improve navigation performance. In addition, a novel tangent path reward (TPR) calculation method is developed to adequately utilize the known contours and positions of obstacles during the training phase. Experimental results demonstrate that the proposed method enables quadrotors to autonomously navigate complex virtual obstacle environments with superior efficiency compared with other algorithms. Furthermore, the feasibility and adaptability of the proposed method are validated through simulations by varying obstacle density and map size, as well as replicating real-world obstacle distributions.}, archive = {J_ESWA}, author = {Qizhang Luo and Yuqi Li and Jiaheng Zeng and Guohua Wu and Yalin Wang}, doi = {10.1016/j.eswa.2025.129762}, journal = {Expert Systems with Applications}, month = {3}, pages = {129762}, shortjournal = {Expert Syst. Appl.}, title = {Quadrotor navigation considering attitude: A deep reinforcement learning method using tangent path rewards}, volume = {298}, year = {2026}, }
• (2026). A dynamic tri-stage framework with neural network-assisted search for constrained multi-objective optimization. ESWA, 298, 129761. (www)

  @article{ , abstract = {Constrained multi-objective optimization problems involve the optimization of multiple objective functions and the satisfaction of different constraints, which poses a challenge for algorithms to achieve a good balance between convergence and diversity. However, indiscriminately enhancing diversity can hinder convergence, while solely focusing on convergence may impair the exploration of the objective space, especially when the current stage is not well-defined. To address this issue, we propose a three-stage multi-task framework for constrained multi-objective optimization with dynamically switchable stages. This framework introduces two auxiliary tasks: one that operates during the exploration and transition stages to accelerate convergence towards the boundary of the infeasible regions and assist the population in crossing it, and another that operates in the final convergence stage to guide the population towards the constrained Pareto front. Moreover, a stage detection method is proposed, which evaluates the current stage to determine the appropriate evolutionary direction for the population, thus enabling dynamic stage transitions. In addition, a neural network-assisted search operator is designed for the auxiliary task during the transition stage, which learns the optimal offspring generation process. This operator enhances the ability of the auxiliary population to cross the infeasible regions. Finally, the performance of the proposed algorithm is superior and competitive on three test suites and six real-world engineering problems compared to seven state-of-the-art algorithms.}, archive = {J_ESWA}, author = {Qianlong Dang and Xinkang Hong and Xianpeng Sun}, doi = {10.1016/j.eswa.2025.129761}, journal = {Expert Systems with Applications}, month = {3}, pages = {129761}, shortjournal = {Expert Syst. Appl.}, title = {A dynamic tri-stage framework with neural network-assisted search for constrained multi-objective optimization}, volume = {298}, year = {2026}, }
• (2026). A physics-informed neural network surrogate model and many-objective optimization algorithm for coupled multi-energy systems in smart grids. ESWA, 298, 129760. (www)

  @article{ , abstract = {The scope of smart grids is progressively extending toward Integrated Energy Systems (IES) that couple electricity with gas, heating, and cooling. Due to the unsteady-state physical characteristics inherent in the transmission of gas, heat, and cooling resources, IES scheduling must not only balance multiple typical objectives but also account for the dynamic coupling of heterogeneous physical domains. To address these challenges, this paper formulates a Many-objective Optimization Model for Coupled Multi-Energy Flows (MaOCMFM) with partial differential equations (PDEs) in IES, which captures the dynamic physical behaviors of electricity, gas, heat, and cooling subsystems. Building upon this model, we propose a Probabilistic Contributing Many-objective Evolutionary Algorithm enhanced by a Physics-Informed Neural Network surrogate model (PC-MaOEA-PINN). Cubic B-spline functions are employed to achieve a continuous representation of the decision variables, while multi-physics constraints are embedded into the loss function of the surrogate model. This design enables efficient approximation of the objective function with a limited number of samples and facilitates focused exploration in critical evolutionary regions, thereby accelerating population convergence. The effectiveness of the proposed model and algorithm is validated on 9 typical scheduling days across four simulated IES scenarios.}, archive = {J_ESWA}, author = {Jingbo Zhang and Xingjuan Cai and Zhihua Cui and Jinjun Chen}, doi = {10.1016/j.eswa.2025.129760}, journal = {Expert Systems with Applications}, month = {3}, pages = {129760}, shortjournal = {Expert Syst. Appl.}, title = {A physics-informed neural network surrogate model and many-objective optimization algorithm for coupled multi-energy systems in smart grids}, volume = {298}, year = {2026}, }
• (2026). An emergency scheduling method based on AutoML for space maneuver objective tracking. ESWA, 298, 129759. (www)

  @article{ , abstract = {The rapid development of large-scale constellations has led to a dramatic increase in the number of Resident Space Objectives (RSOs), significantly intensifying the complexity of Space Situational Awareness (SSA). Furthermore, the maneuver behaviors of non-cooperative RSOs pose potential threats to space safety, making the real-time monitoring of their post-maneuver orbital becomes more critical. In particular, the maneuvering characteristics of large-scale constellation satellites impose more stringent demands on the timeliness and adaptability of existing scheduling algorithms for observation resources. To address the emergency scheduling demands of heterogeneous ground-based observation resources, this paper proposes an Emergency Task Three-phase Scheduling Framework (ETTSF) based on Automated Machine Learning (AutoML) and auction algorithm. The framework collaboratively optimizes resource allocation through three phases: resource matching, task scheduling, and rescheduling. First, AutoML combined with an auction algorithm predicts and assigns emergency tasks to the most appropriate resources, reducing solution space complexity, simultaneously, the auction algorithm’s corrected results are fed back to AutoML for model fine-tuning. Second, a heuristic algorithm with dynamic neighborhood structures efficiently inserts emergency tasks into the routine observation plan. Finally, affected routine tasks are rescheduled to minimize the operational impact. Simulation results demonstrate that compared to baselines-Real-Time Dynamic Scheduling (RTDS) and Improved Adaptive Large Neighborhood Search (IALNS)-ETTSF achieves a 2.82% improvement in Completion Rate of Emergency RSOs (CRER) and a 27.83% reduction in Impact Rate (IR) on routine tasks. Ablation experiments further validate the effectiveness of the resource matching and rescheduling phases.}, archive = {J_ESWA}, author = {Xi Long and Jinrun Chen and Leping Yang and Huan Huang}, doi = {10.1016/j.eswa.2025.129759}, journal = {Expert Systems with Applications}, month = {3}, pages = {129759}, shortjournal = {Expert Syst. Appl.}, title = {An emergency scheduling method based on AutoML for space maneuver objective tracking}, volume = {298}, year = {2026}, }
• (2026). Path-aware routing system for multimodal vigilance estimation: A structured fusion perspective. ESWA, 298, 129757. (www)

  @article{ , abstract = {Driver vigilance is a critical cognitive factor in ensuring the safety of intelligent driving systems. With advances in physiological and behavioral sensing technologies, multimodal data have become an essential source for modeling drivers’ cognitive states. However, vigilance, as an implicit cognitive state, is difficult to model accurately using a single modality. Efficient fusion of multiple modalities and structured information interaction remains a core challenge in this task. Existing approaches often rely on strategies such as feature concatenation and attention mechanisms, which lack explicit structural constraints. As a result, heterogeneous modality features are fused in an uncontrolled and entangled manner, making the interaction process opaque and difficult to interpret. To address these issues, we propose the Path-Aware Routing System (PARS), which formulates multimodal fusion as a feature routing task. In PARS, intra-modal enhancement and cross-modal interaction are abstracted into independent semantic channels, and a confidence-aware mechanism is introduced to enable dynamically weighted fusion. PARS explicitly constructs a path space, allowing features to be selectively routed and integrated along semantical pathways, thereby enhancing the model’s discriminative power and robustness. We conduct extensive experiments on a public dataset and a self-constructed driving simulation dataset. The results demonstrate that PARS significantly outperforms existing multimodal fusion methods in the task of driver vigilance estimation, achieving superior performance in terms of accuracy, interpretability, and generalization. These findings highlight the broad potential of PARS in intelligent driving applications. Our code and models are available at: https://github.com/SunYu-Gavin/PARS .}, archive = {J_ESWA}, author = {Yu Sun and Shiwu Li and Yiming Bie and Linhong Wang and Tongtong Jin and Mengzhu Guo and Zhifa Yang}, doi = {10.1016/j.eswa.2025.129757}, journal = {Expert Systems with Applications}, month = {3}, pages = {129757}, shortjournal = {Expert Syst. Appl.}, title = {Path-aware routing system for multimodal vigilance estimation: A structured fusion perspective}, volume = {298}, year = {2026}, }
• (2026). Homophone-aware offensive language detection via semantic-phonetic collaboration. ESWA, 298, 129756. (www)

  @article{ , abstract = {The increasing use of implicit and obfuscated expressions poses significant challenges to offensive language detection in Chinese online platforms. In particular, users often exploit homophone substitutions to bypass keyword-based moderation, making traditional detection systems inadequate. This study addresses the problem of detecting offensive content masked through homophonic substitutions, which retain aggressive intent while altering character representations. Existing methods fall into two main categories: (1) semantic-only models, which struggle with phonetic manipulations due to their reliance on text features alone, and (2) auxiliary-enhanced models, which incorporate phonetic or syntactic signals but lack deep integration between modalities. To overcome these limitations, we propose a lightweight dual-branch model that separately encodes textual semantics and pinyin phonetics under a multi-view learning framework. A Dual-Branch Interactive Training strategy is introduced to enable dynamic cross-modal alignment via contrastive objectives, allowing each modality to mutually refine the other and enhance robustness to adversarial inputs. We conduct experiments on two benchmark datasets, COLD and SWSR, both of which are augmented with varying levels of homophone noise to simulate real-world evasion strategies. The proposed model outperforms all baseline models, achieving an average F1-score improvement of 6.3 % under high-noise conditions, while reducing inference latency and memory usage by more than 60 %, demonstrating both effectiveness and efficiency for real-time deployment. We will release the source code for further use by the community https://github.com/hjhhlc/DBIT .}, archive = {J_ESWA}, author = {Jiahao Hu and Shanliang Pan}, doi = {10.1016/j.eswa.2025.129756}, journal = {Expert Systems with Applications}, month = {3}, pages = {129756}, shortjournal = {Expert Syst. Appl.}, title = {Homophone-aware offensive language detection via semantic-phonetic collaboration}, volume = {298}, year = {2026}, }
• (2026). Enhancing traffic signal control through model-based reinforcement learning and policy reuse. ESWA, 298, 129755. (www)

  @article{ , abstract = {Multi-agent reinforcement learning (MARL) has shown significant potential in traffic signal control (TSC). However, current MARL-based methods often suffer from insufficient generalization due to the fixed traffic patterns and conditions of the road network used during training. This limitation results in poor adaptability to new traffic scenarios, leading to high retraining costs and complex deployment. To address this challenge, we propose two algorithms: PLight and PRLight. PLight employs a model-based reinforcement learning approach, pretraining control policies, and environment models using predefined source-domain traffic scenarios. The environmental model predicts state transitions, facilitating the comparison of environmental characteristics. PRLight further enhances adaptability by adaptively selecting pre-trained PLight agents based on the similarity between the source and target domains to accelerate the learning process in the target domain. We evaluated the algorithms through two transfer settings: (1) adaptability to different traffic scenarios within the same road network, and (2) generalization across different road networks. The results show that PRLight significantly reduces the adaptation time compared to learning from scratch in new TSC scenarios, achieving optimal performance using similarities between available and target scenarios.}, archive = {J_ESWA}, author = {Yihong Li and Chengwei Zhang and Furui Zhan and Wanting Liu and Kailing Zhou and Longji Zheng}, doi = {10.1016/j.eswa.2025.129755}, journal = {Expert Systems with Applications}, month = {3}, pages = {129755}, shortjournal = {Expert Syst. Appl.}, title = {Enhancing traffic signal control through model-based reinforcement learning and policy reuse}, volume = {298}, year = {2026}, }
• (2026). Federated transfer learning for anomaly detection in HPC systems: First real-world validation on a tier-0 supercomputer. ESWA, 298, 129754. (www)

  @article{ , abstract = {High-Performance Computing (HPC) systems increasingly require intelligent, scalable anomaly detection to ensure operational reliability. However, conventional centralized approaches often struggle with data privacy constraints, poor generalization across heterogeneous nodes, and limited scalability. This study presents the first real-world application of federated transfer learning (FTL) for anomaly detection in a production-grade Tier-0 supercomputer. By combining federated learning with transfer learning, the proposed framework enables decentralized model training and personalized adaptation to unseen nodes, without accessing raw data. We validate the approach using two large-scale telemetry datasets collected from 100 nodes of the Marconi100 supercomputer, evaluating its effectiveness across supervised, semi-supervised, and unsupervised learning paradigms. Results show that FTL consistently improves anomaly detection performance on nodes that did not participate in federated training, with F1-score gains reaching up to 0.50. These improvements demonstrate the framework’s ability to generalize across non-identically distributed data and maintain detection accuracy under real-world conditions. This work establishes FTL as a scalable, privacy-preserving solution for fault detection in HPC environments. Its practical deployment on production hardware confirms its readiness for real-time monitoring applications in large-scale, heterogeneous computing systems.}, archive = {J_ESWA}, author = {Emmen Farooq and Michela Milano and Andrea Borghesi}, doi = {10.1016/j.eswa.2025.129754}, journal = {Expert Systems with Applications}, month = {3}, pages = {129754}, shortjournal = {Expert Syst. Appl.}, title = {Federated transfer learning for anomaly detection in HPC systems: First real-world validation on a tier-0 supercomputer}, volume = {298}, year = {2026}, }
• (2026). Security script arrangement based on enhanced BERT for cooperative defense in networked control systems. ESWA, 298, 129753. (www)

  @article{ , abstract = {Due to the mutual collaboration and in-depth integration among multiple defense technologies through information sharing, the cooperative defense in networked control systems has emerged as a feasible solution to counter increasingly diversified cyber threats under the unique security characteristics and requirements of industrial environments. However, one of the chief challenges is how to automatically and intelligently develop effective cooperative working strategies when an attack occurs. Leveraging the advantages of large-scale AI (Artificial Intelligence) models, this paper defines a new concept named “security script”, and proposes a security script arrangement approach based on enhanced BERT to achieve fine-grained cooperative defense in networked control systems. Furthermore, this approach introduces intrusion detection and industrial firewall as two practical examples, and can automatically arrange effective security scripts to enable the dynamic interaction of two defense technologies. Additionally, to improve efficiency, the encoder structure adjusting and AdamW optimizing are further presented to enhance the traditional BERT. Experimental results clearly demonstrate that: for one thing, these two optimization ways can make greater achievements in reducing unnecessary time consumption and enhancing accuracy of security script arrangement; for another, compared with other typical BERT and large-scale AI models, the proposed approach can exhibit more favorable performance advantages in achieving cooperative defense based on security script arrangement. In particular, through its successful application and verification in one real-world manufacturing control system, our approach may bring a potential opportunity or direction for further research and improvement of AI-based cooperative defense.}, archive = {J_ESWA}, author = {Ming Wan and Xueqing Liu and Shengbao An and Aiping Tan and Xi Jin and Chuan Sheng}, doi = {10.1016/j.eswa.2025.129753}, journal = {Expert Systems with Applications}, month = {3}, pages = {129753}, shortjournal = {Expert Syst. Appl.}, title = {Security script arrangement based on enhanced BERT for cooperative defense in networked control systems}, volume = {298}, year = {2026}, }
• (2026). ScSCDT: Self-contrastive neural network with deep topology mining for scRNA-seq data clustering. ESWA, 298, 129751. (www)

  @article{ , abstract = {Advancements in single-cell sequencing technologies have enabled researchers to better identify cells based on gene-level information. Cell clustering is a key task in single-cell analysis and plays an important role in distinguishing cell types. However, due to the high dimensionality and sparsity of scRNA-seq data, single-cell clustering remains a major challenge. Although many methods based on deep learning and machine learning have been developed for single-cell clustering, they often fail to capture the deep topological structure between cells, which limits clustering precision. In addition, most existing clustering approaches cannot effectively construct suitable sample pairs to optimize clustering models. To address these issues, we propose a topology-aware deep contrastive clustering model for single-cell data, named scSCDT. First, scSCDT employs a ZINB-based autoencoder to simultaneously learn cell embeddings and topological information, effectively handling the challenges posed by the high-dimensional and sparse nature of the data. Then, we introduce a dual clustering-guided loss to supervise the clustering task, combining a probabilistic soft assignment strategy and a hard pseudo-labeling strategy for optimization. Finally, based on the topological structure in the low-dimensional embedding space, we construct negative pairs within a single view and design a self-contrastive learning method to further improve clustering performance. We conduct extensive experiments on ten real scRNA-seq datasets and evaluate performance using four clustering metrics. The results indicate that scSCDT achieves strong clustering performance across multiple datasets, thereby facilitating more accurate cell type identification in single-cell transcriptomic analysis.}, archive = {J_ESWA}, author = {Zhongyang Zhou and Bin Tang and Feiyu Chen and Wei Wang and Shangshang Zhao and Nanjun Yu}, doi = {10.1016/j.eswa.2025.129751}, journal = {Expert Systems with Applications}, month = {3}, pages = {129751}, shortjournal = {Expert Syst. Appl.}, title = {ScSCDT: Self-contrastive neural network with deep topology mining for scRNA-seq data clustering}, volume = {298}, year = {2026}, }
• (2026). 3D modeling from a single sketch with multifaceted semantic understanding. ESWA, 298, 129748. (www)

  @article{ , abstract = {This paper studies the problem of 3D shape generation from a single sketch. Prior works rely on directly extracted visual features of sketches as guidance for the generation process. However, the sparse visual cues and abstract nature of sketches, which are inherited in the guiding features, lead to semantic ambiguity and geometry incompleteness in the generated shapes, compromising accuracy. To address this, we propose MSU-3D, a diffusion-based framework for sketch-to-3D generation, leveraging Multifaceted Semantic Understanding to explicitly analyze the construction information of sketches from multiple facets before providing fine-grained guidance over 3D shape generation. Specifically, we decompose sketches through three interpretative facets (semantics, depth, and normal), introducing reasoning of three representations to capture 3D features from distinct perspectives: local components, basic 3D geometry, and 3D surface details. One step further, we propose a multifaceted perception module. It aggregates multifaceted feature representations and leverages local component features as a two-pronged guiding representation to jointly guide the perception of basic shapes and surface details. To ensure fine-grained control, the hierarchical perception strategy adaptively injects varying granularity of perception features at different stages of the 3D generation. Extensive experiments and comparisons with state-of-the-art methods on various complex posture datasets validate the effectiveness of our framework in mitigating semantic ambiguity and geometry incompleteness in 3D generation.}, archive = {J_ESWA}, author = {Yuxiao Zhang and Jin Wang and Yang Zhou and Senyun Jia and Zhi Zheng and Dongliang Zhang and Guodong Lu}, doi = {10.1016/j.eswa.2025.129748}, journal = {Expert Systems with Applications}, month = {3}, pages = {129748}, shortjournal = {Expert Syst. Appl.}, title = {3D modeling from a single sketch with multifaceted semantic understanding}, volume = {298}, year = {2026}, }
• (2026). Continuous learning approach to synergize shimmering image enhancement and de-fogging with small sample. ESWA, 298, 129747. (www)

  @article{ , abstract = {Shimmering Image Enhancement and Defogging (SIED) are two important aspects of image recovery. However, most methods are often fail to consider image context information, overexposure of image and the intrinsic correlation between SIED without enough sample. Furthermore, current methods may lead to the amplification of external color interference during image recovery, and can’t fine-tune the model with new samples. Firstly, we propose an Adjustable Multiscale Attention Codec Network (AMACNet) architecture. AMACNet includes variable restorative coder decoder and group channel attention to fuse multi-level contextual and channel information of images, respectively. These components are designed as plug-and-play modules. Secondly, a Continuous Learning with Small Sample (CLS) training method is presented. It utilizes few samples based on a front-and-back stage receding structure. The method synergizes the tasks of SIED, allowing the network to perform both tasks without incresing the number of parameters. It also enables the network to adapt to new tasks with a small number of unpaired samples. Finally, a color adjustment module for image post-processing is designed. The post-processing method is used to balance the impact of color illumination on the inherent color of materials. Extensive experiments demonstrate that training AMACNet by CLS allow the final image recovery results to exceed the GT in terms of ENIQA, FES, PIQE, and other unparameterized metrics.}, archive = {J_ESWA}, author = {Fenglin Yao and Zhengxiang Liu}, doi = {10.1016/j.eswa.2025.129747}, journal = {Expert Systems with Applications}, month = {3}, pages = {129747}, shortjournal = {Expert Syst. Appl.}, title = {Continuous learning approach to synergize shimmering image enhancement and de-fogging with small sample}, volume = {298}, year = {2026}, }
• (2026). Learning domain-invariant representation for generalizable iris segmentation. ESWA, 298, 129746. (www)

  @article{ , abstract = {Cross-domain iris segmentation (CDIS) seeks to transfer knowledge from a labeled source dataset to an unlabeled target dataset. Existing CNN-based iris segmentation methods commonly assume that training and application stages share the same data distribution and modality setting, thus their performance may decline substantially on open-domain iris datasets unseen before. Furthermore, the process of annotating pixel-wise labels is labor-intensive and time-consuming, resulting in limited applicability of these methods in realistic scenarios. Therefore, we propose a generic domain adaptation iris segmentation framework ( DAIrisSeg ), which can be flexibly incorporated into existing methods. First, a domain-sensitive feature whitening strategy is proposed to effectively mitigate the domain-specific styles while preserving the domain-invariant content, thereby improving the model’s generalizability to unknown domain distribution. We then utilize the prototype estimation and the context-similarity learning adapter to produce reliable segmentation labels. In addition, DAIrisSeg incorporates prior constraints of the iris to further refine the segmentation results. Extensive experiments on three iris datasets demonstrate that the proposed method has shown consistent improvements over state-of-the-art (SOTA) methods.}, archive = {J_ESWA}, author = {Dawei Lin and Meng Yuan and Ying Chen and Xiaodong Zhu and Yuanning Liu}, doi = {10.1016/j.eswa.2025.129746}, journal = {Expert Systems with Applications}, month = {3}, pages = {129746}, shortjournal = {Expert Syst. Appl.}, title = {Learning domain-invariant representation for generalizable iris segmentation}, volume = {298}, year = {2026}, }
• (2026). Cycle-CFM: An unsupervised framework for robust multimodal anomaly detection in industrial settings. ESWA, 298, 129745. (www)

  @article{ , abstract = {Industrial multimodal anomaly detection is confronted with three pivotal challenges: cross-modal feature drift, noise sensitivity, and modality imbalance. To address these issues, we propose Cycle-Consistent Cross-Modal Feature Mapping (Cycle-CFM), an unsupervised framework that integrates cycle-consistent cross-modal mapping with channel-attention-guided adaptive loss weighting. Cycle-CFM establishes bidirectional feature alignment between RGB and 3D modalities via reversible cycle mappings, yielding consistent representations robust to vibration and depth noise. To further mitigate dynamic interferences such as illumination variations, we introduce a joint optimization strategy that combines cross-consistency and cycle-consistency losses. Experimental results on our self-constructed SteelDefect-3D-AD dataset demonstrate that Cycle-CFM achieves an AUPRO@1 % of 0.371, outperforming state-of-the-art methods by 17–45 %. It also attains a pixel-level AUROC (P-AUROC) of 0.991 and an image-level AUROC (I-AUROC) of 0.998. On the public MVTec 3D-AD benchmark, Cycle-CFM reaches a mean P-AUROC of 0.960 and improves accuracy by 37.5 % for elongated anomalies. With a runtime of 11.03 FPS and 469.52 MB of parameters, the model highlights both its effectiveness and deployability for real-time industrial inspection.}, archive = {J_ESWA}, author = {Yikang Shi and Xin Zhan and Yaqian Li and Zhongqiang Wu and Wenming Zhang and Haibin Li}, doi = {10.1016/j.eswa.2025.129745}, journal = {Expert Systems with Applications}, month = {3}, pages = {129745}, shortjournal = {Expert Syst. Appl.}, title = {Cycle-CFM: An unsupervised framework for robust multimodal anomaly detection in industrial settings}, volume = {298}, year = {2026}, }
• (2026). An explainable transfer learning approach to predict carbon emission intensity of coal-fired power plants with multi-source monitoring data. ESWA, 298, 129743. (www)

  @article{ , abstract = {Carbon mitigation policies and emission trading systems have heightened the need to monitor and predict the carbon emission intensity (CEI) of coal-fired power plants. Leveraging big data and machine learning (ML) technologies, this study trains predictive models of CEI using operational parameters, load rate, and coal-quality data from three Chinese power plants. The performance of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), and Artificial Neural Network (ANN) was evaluated, and the challenge of limited data in individual plants was mitigated through instance-based transfer learning (ITL) and graft learning (GL) technologies. The results indicate that while traditional ML models struggle with poor data quality and limited samples, transfer learning between different plants will improve predictive accuracy substantially, and GL delivers the greatest gains. Among the most influential features, air supply temperature and load rate critically impact CEI and therefore should be carefully managed to achieve emission reductions. Nevertheless, the effectiveness of transfer learning depends on source data quality and model choice, and the proposed operational strategies require further validation before practical adoption. Our findings offer a robust framework for enhancing the predictive accuracy of plant CEI under data-scarce conditions and inform effective strategies for promoting a low-carbon transition in the energy sector.}, archive = {J_ESWA}, author = {Xiaodong Jin and Lingzhen Zhang and Fangyi Li and Wu Xie and Dawei Ma and Yan Wu}, doi = {10.1016/j.eswa.2025.129743}, journal = {Expert Systems with Applications}, month = {3}, pages = {129743}, shortjournal = {Expert Syst. Appl.}, title = {An explainable transfer learning approach to predict carbon emission intensity of coal-fired power plants with multi-source monitoring data}, volume = {298}, year = {2026}, }
• (2026). Partially view-aligned clustering via data recoupling and elastic bi-consistency learning. ESWA, 298, 129741. (www)

  @article{ , abstract = {Modern multi-view data often suffer from partial view alignment issues, yet most existing multi-view clustering (MVC) methods assume that the data are fully aligned, which is rarely the case in real-world scenarios. This misalignment leads to False Negative Pairs (FNPs), disrupting the learning process. While some methods address partial alignment, they often neglect intra-view consistency and multi-scale inter-view relationships, limiting their ability to capture both global and local structural dependencies. Additionally, the prevalent use of Mean Squared Error (MSE) as a reconstruction loss is suboptimal for discrete data, potentially causing severe performance degradation. To overcome these limitations, we propose Partially View-aligned Clustering via Data Recoupling and Elastic Bi-consistency Learning (PVC-DREBL). Our method integrates two key components: (1) a Data Recouple Module, which realigns the data to mitigate the effects of FNPs while leveraging an exponential contrastive loss to enhance learning stability and prevent overfitting; (2) an Elastic Bi-consistency Learning Module, designed to reconstruct diverse data types robustly while enforcing intra-view and multi-scale inter-view consistency. Extensive experiments on six benchmark datasets demonstrate that PVC-DREBL significantly outperforms existing methods, highlighting its effectiveness in handling partially view-aligned clustering tasks.}, archive = {J_ESWA}, author = {Wenzhe Liu and Jiongcheng Zhu and Jingbo Tan and Huibing Wang and Yong Zhang}, doi = {10.1016/j.eswa.2025.129741}, journal = {Expert Systems with Applications}, month = {3}, pages = {129741}, shortjournal = {Expert Syst. Appl.}, title = {Partially view-aligned clustering via data recoupling and elastic bi-consistency learning}, volume = {298}, year = {2026}, }
• (2026). Attention-based CNN for lithological identification from logging curves: A case study in the qaidam basin, china. ESWA, 298, 129740. (www)

  @article{ , abstract = {Subsurface lithological distribution is essential for extrapolating geological information from core to block or basin scales. Given the limited availability of core data, there is a critical need to develop a reliable method for establishing robust correlations between logging curves and lithologies in cores, thereby maximizing the value of large historical logging data. Here, we propose a novel attention-based convolutional neural network (ATT-CNN), which employs a 1D-CNN to transform six types of logging curves into high-dimensional feature space at each depth, and applies an attention mechanism to the 1D-CNN outputs along both the depth and feature dimensions. The architecture is designed to mimic human perceptual processing for lithology identification, leveraging curve combination, thresholding, and local pattern recognition within this enriched and high-dimensional feature representation. In addtion, the study employs wavelet-based preprocessing on logging curves to eliminate the impact of compaction-induced data drift on model generalization—an issue rarely considered in prior studies. The result showes that: ① The proposed ATT-CNN model demonstrates superior performance over benchmark models—the bidirectional gated recurrent unit (BiGRU) and an ensemble of machine learning models (En-ML)—across all evaluation metrics; ②Wavelet-based preprocessing enhances the generalization capability of both ATT-CNN and BiGRU, yielding higher metric scores and improved predictions, particularly in shallow-depth intervals; ③ For blind wells, the ATT-CNN outperforms BiGRU and En-ML in both accuracy and its ability to capture lithological variations even from low-amplitude curve deviations. The integration of ATT-CNN with wavelet-based preprocessing demonstrates significant potential for accurately characterizing subsurface lithological distribution, then provides critical support for key petroleum geology workflows, including provenance analysis, sedimentary facies mapping, and reservoir property prediction.}, archive = {J_ESWA}, author = {Jianguo Yin and Shuai Zhang and Zhixiong Wu and Shouji Pang and Rui Wang}, doi = {10.1016/j.eswa.2025.129740}, journal = {Expert Systems with Applications}, month = {3}, pages = {129740}, shortjournal = {Expert Syst. Appl.}, title = {Attention-based CNN for lithological identification from logging curves: A case study in the qaidam basin, china}, volume = {298}, year = {2026}, }
• (2026). Clustering-based brain functional segmentation via deep collapsed nonparametric von mises-fisher mixture models. ESWA, 298, 129739. (www)

  @article{ , abstract = {This paper presents a clustering-based framework for the analysis of functional magnetic resonance imaging (fMRI) data, with a particular focus on brain segmentation into functional sub-regions. The proposed approach comprises two key modules: representation learning and brain functional segmentation. To extract meaningful latent representations from high-dimensional fMRI signals while preserving temporal dependencies, we introduce the Spherical Variational Recurrent Autoencoder (SVRAE), a deep generative model built upon the Variational Autoencoder (VAE) architecture. Unlike conventional VAEs that assume a Gaussian prior, SVRAE employs the von Mises-Fisher (vMF) distribution to model latent variables on a unit hypersphere, which is more suitable for L 2 -normalized data. To further enhance temporal modeling, we replace standard fully connected layers with Long Short-Term Memory (LSTM) networks. For the segmentation module, we adopt a Collapsed Nonparametric von Mises-Fisher Mixture Model (Co-vMFMM), formulated within a Bayesian nonparametric framework. This model automatically adapts its complexity to the input data without requiring a predefined number of clusters. An efficient variational Bayes learning algorithm is developed to perform inference in a collapsed parameter space. Extensive experiments on publicly available fMRI datasets demonstrate the effectiveness and robustness of the proposed method in delineating functionally coherent brain sub-regions.}, archive = {J_ESWA}, author = {Wentao Fan and Wenchuan Zhang and Xiao Dong and Nizar Bouguila}, doi = {10.1016/j.eswa.2025.129739}, journal = {Expert Systems with Applications}, month = {3}, pages = {129739}, shortjournal = {Expert Syst. Appl.}, title = {Clustering-based brain functional segmentation via deep collapsed nonparametric von mises-fisher mixture models}, volume = {298}, year = {2026}, }
• (2026). An optimized hierarchical path planning method based on deep reinforcement learning for mobile robots. ESWA, 298, 129736. (www)

  @article{ , abstract = {With the development of elderly mobile robot technology, the training efficiency and path planning of robots have become key issues in research. Current mobile robot training faces challenges such as local optima and slow convergence speeds. To address these issues, this paper proposes a hierarchical path planning method based on deep reinforcement learning (H-DDQN). This method first introduces a global path planning module, which uses a density function to select high-value key points. In areas with dense obstacles, these key points provide the robot with effective global path guidance, enabling it to avoid getting stuck in a local optimum due to reliance solely on local information. In the local path planning module, we introduce a spatial attention mechanism based on global information, which weights global path points to enable the robot to focus on critical areas, thereby enhancing local decision-making capabilities and addressing dynamic obstacles on the basis of global optimality. Finally, this paper designs a new comprehensive reward function that combines path guidance from global key points with goal-oriented dense rewards, avoiding excessive unnecessary exploration and providing timely feedback to accelerate the model’s convergence speed. Experimental results show that compared to other existing algorithms, the H-DDQN algorithm converges more quickly during training and generates shorter and more efficient paths. In dynamic obstacle environments, the algorithm also demonstrates strong adaptability, combining global and local capabilities to achieve superior performance.}, archive = {J_ESWA}, author = {Jilin Zhang and Jia Qiao and Ke Huang and Menghua Zhang}, doi = {10.1016/j.eswa.2025.129736}, journal = {Expert Systems with Applications}, month = {3}, pages = {129736}, shortjournal = {Expert Syst. Appl.}, title = {An optimized hierarchical path planning method based on deep reinforcement learning for mobile robots}, volume = {298}, year = {2026}, }
• (2026). Assessing the renewable energy sources for sustainable energy generation systems: Interval-valued q-rung orthopair fuzzy SWARA-TOPSIS. ESWA, 298, 129735. (www)

  @article{ , abstract = {Renewable Energy Sources (RESs) help decarbonize power systems, but selecting among them is a challenging decision problem due to multiple, often conflicting, technical, economic, environmental, and health-related criteria. Consequently, numerous studies in the literature have attempted to address this decision-making issue using objective, subjective, and fuzzy decision-making procedures. However, there are still unaddressed research gaps in the literature, particularly regarding the explicit modeling of expert hesitation and ambiguity in real-world RES selection cases. The current study develops a decision-making model based on Step-wise Weight Assessment Ratio Analysis (SWARA) and Technique of Order Preference Similarity to the Ideal Solution (TOPSIS) methods integrated with Interval-Valued q-Rung Orthopair Fuzzy Sets (IV-q-ROFSs) to fill these gaps. Unlike previous studies that have predominantly applied conventional fuzzy MCDM techniques, our model introduces the first integration of IV-q-ROFS into RES selection. This novelty enables a more accurate representation of expert hesitation and uncertainty. The study is applied to a real industrial case in Turkey, where six RES alternatives are evaluated across 43 criteria by five senior experts under the supervision of a three-member professionals’ board. Furthermore, the structured robustness check and systematic literature mapping ensure that the proposed approach is methodologically robust and practically relevant for policymakers and energy planners. The application results of the developed model demonstrate that the estimated energy production potential of the RES and the effects of carcinogens generated from utilizing these energy sources are the critical factors influencing the selection of the most appropriate RESs. Solar energy ranked first among the alternatives. The applicability and validity of the developed model are examined by a comprehensive robustness check consisting of tests of sensitivity, comparison, and resilience to the rank reversal problem. Overall, the study provides (i) a novel methodological framework integrating IV-q-ROFS with SWARA and TOPSIS, (ii) empirical evidence from a comprehensive real-world RES selection case, and (iii) policy-relevant insights into the drivers of renewable energy adoption.}, archive = {J_ESWA}, author = {Ömer Faruk Görçün and Ahmet Aytekin and Selçuk Korucuk and Erfan Babaee Tirkolaee}, doi = {10.1016/j.eswa.2025.129735}, journal = {Expert Systems with Applications}, month = {3}, pages = {129735}, shortjournal = {Expert Syst. Appl.}, title = {Assessing the renewable energy sources for sustainable energy generation systems: Interval-valued q-rung orthopair fuzzy SWARA-TOPSIS}, volume = {298}, year = {2026}, }
• (2026). Enhancing liver cancer detection: An innovative deep learning approach combining GAN, ResNet, and vision transformer. ESWA, 298, 129734. (www)

  @article{ , abstract = {Liver cancer is a complex and life-threatening disease with significant diagnostic and therapeutic challenges. Automated liver cancer detection assists radiologists in identifying tumors and their severity accurately. In recent years, several deep-learning techniques have been implemented for diagnosing liver tumors and classification. Despite advancements in deep learning for medical imaging, existing liver cancer detection approaches continue to face several critical limitations. These include suboptimal diagnostic accuracy due to inadequate feature extraction, excessive computational demands that hinder real-time deployment, significant class imbalance within medical datasets leading to biased predictions, and overfitting caused by limited annotated training data. To address these challenges, this study introduces a novel and automated deep learning framework called CustomLiverNet, specifically designed for accurate and efficient liver cancer diagnosis using Computed Tomography images. The Generative Adversarial Network is introduced for generating realistic synthetic images, effectively improving the performance of the proposed technique and reducing class imbalance problems. The proposed technique integrates the strengths of Residual Networks and Vision Transformer to extract significant information from the input images and further enhance the performance of the proposed framework. The Residual Networks capture both low-level and high-level semantic features, whereas the Vision Transformer derives global and contextual feature representations from the input images. The model designs a customized fusion layer for combining the extracted features from both Residual Networks and Vision Transformer models. The classification layer predicts whether the liver tumor is benign or malignant. Further, Gradient-Weighted Class Activation Mapping is applied to highlight the critical regions of the image to enhance model transparency. The CustomLiverNet framework was trained and validated on two publicly available liver cancer datasets, including the liver tumor segmentation dataset, which contains 131 contrast-enhanced abdominal Computed Tomography scans, and the 3D image reconstruction for comparison of algorithm database, which includes 20 computed tomography scans. Experimental evaluation using standard metrics shows that CustomLiverNet achieved an accuracy of 98.79 %, precision of 98.64 %, recall of 98.58 %, and specificity of 98.35 %. These results demonstrate that the proposed model holds strong potential for enhancing early and accurate liver cancer diagnosis compared to previous studies.}, archive = {J_ESWA}, author = {Shivani Joshi and Avinash Dwivedi and Rajiv Kumar and Ashish Kumar and Raju Kumar and Amrita}, doi = {10.1016/j.eswa.2025.129734}, journal = {Expert Systems with Applications}, month = {3}, pages = {129734}, shortjournal = {Expert Syst. Appl.}, title = {Enhancing liver cancer detection: An innovative deep learning approach combining GAN, ResNet, and vision transformer}, volume = {298}, year = {2026}, }
• (2026). Unleashing the power of large language models for low-resource relation triplet extraction by structure-to-text data generation. ESWA, 298, 129733. (www)

  @article{ , abstract = {Scaling language models have revolutionized widespread NLP tasks, yet little investigation has been conducted to assess the ability of Large Language Models (LLMs) to explore low-resource relation triplet extraction. This paper investigates essential methodologies, k -shot demonstration of in-context learning, and many-shot instruction tuning for few-shot and zero-shot relation triplet extraction using FlanT5, supported by exhaustive experiments. To enhance low-resource setting performance, we further propose different types of demonstration examples and task-related instructions for data generation. Specifically, we leverage LLMs to construct a structured prompt template for generating synthetic training data based on structured text and efficiently explore the boundary issues of examples in both instruction tuning and in-context learning, assessing their impact on model performance. To address the challenge of extracting multiple relation triplets from a single sentence, we design a novel Multiple Triplet Search (MTS) algorithm. Furthermore, we find that in-context learning can match the performance of previous prompt learning methods. Additionally, integrating synthetic data with the LLM can improve existing solutions in low-resource scenarios, achieving new state-of-the-art results. Experiments conducted on six relation extraction datasets demonstrate the efficacy of the proposed model for the zero-shot and few-shot RTE tasks. Our source code is publicly available at https://github.com/Phevos75/LLMRTE.}, archive = {J_ESWA}, author = {Qian Guo and Yi Guo and Jin Zhao}, doi = {10.1016/j.eswa.2025.129733}, journal = {Expert Systems with Applications}, month = {3}, pages = {129733}, shortjournal = {Expert Syst. Appl.}, title = {Unleashing the power of large language models for low-resource relation triplet extraction by structure-to-text data generation}, volume = {298}, year = {2026}, }
• (2026). LDATA-net: Dynamic feature adaptation for efficient feature learning in resource-limited UAV detection. ESWA, 298, 129725. (www)

  @article{ , abstract = {Unmanned Aerial Vehicle (UAV) image analysis faces the dual challenges of complex background interference and limited onboard computational resources, particularly when processing extreme scale variations across multiple viewpoints. Existing approaches typically enhance detection accuracy by increasing model complexity, but this often leads to parameter proliferation that exceeds the deployment limits of airborne platforms. To address this fundamental contradiction, we propose LDATA-Net (Lightweight Dynamic Aggregation Task-Aligned Network), which pioneers a “Dynamic Feature Adaptation” design paradigm aimed at achieving synergistic optimization between parameter efficiency and detection accuracy. This framework systematically realizes end-to-end dynamic adaptive capabilities through three core components that operate collaboratively across feature extraction, fusion, and detection stages: (1) Dynamic Multi-Branch Depthwise Block (DMBD-Block), whose core innovation is our proposed novel operator DIDWConv, which adaptively adjusts receptive fields according to input features to capture targets of extreme scales and orientations; (2) Lightweight Dynamic Aggregation Network (LDANet), which effectively preserves critical spatial contextual information through hierarchical fusion architecture and dynamic weighting mechanisms; (3) Dynamic Adaptive Head (DA-Head), which effectively mitigates task conflicts through geometric and semantic dynamic feature alignment. LDATA-Net achieves 35.4 %, 77.9 %, and 51.2 % AP 50 on VisDrone2019, DOTA1.0, and AI-TODv2 datasets respectively with only 2.8M parameters, establishing a new paradigm for designing memory-efficient yet high-performance detection systems, particularly for resource-constrained heterogeneous computing aviation platforms.}, archive = {J_ESWA}, author = {Shuming Lin and Sang Feng and Junnan Tan}, doi = {10.1016/j.eswa.2025.129725}, journal = {Expert Systems with Applications}, month = {3}, pages = {129725}, shortjournal = {Expert Syst. Appl.}, title = {LDATA-net: Dynamic feature adaptation for efficient feature learning in resource-limited UAV detection}, volume = {298}, year = {2026}, }
• (2026). Exploring intra- and inter-organizational collaboration opportunities across a technological knowledge ecosystem: A metapath2vec approach. ESWA, 298, 129724. (www)

  @article{ , abstract = {Due to the increasing complexity and scale of technological knowledge ecosystems, organizations face challenges in identifying intra- and inter-organizational collaboration opportunities. In this respect, prior studies have proposed patent-based approaches, but they are subject to several limitations: (1) insufficient consideration of technological relationships within the ecosystem, (2) simplified unit of analysis, and (3) limited organization-centric assessments. This study proposes a network embedding and text-reranking approach to explore potential intra- and inter-organizational collaboration opportunities. First, the technological knowledge ecosystem is represented as a heterogeneous patent network comprising patents, inventors, assignees, and technology classification codes. Second, inventor nodes are embedded using metapath2vec, which performs random walks along predefined metapaths to capture diverse knowledge flows within the ecosystem. Third, potential collaborators are explored through (1) screening candidates based on technological reachability, which measures the possibility of knowledge exploration based on contextual similarity within the network, and (2) reranking candidates based on technological relevance, which quantifies the possibility of knowledge exploitation based on the similarity of technological know-how and experiences. Finally, ten quantitative patent indicators are developed to assess the implications of these opportunities at both the inventor and organization levels. The validity of the proposed approach is demonstrated through a case study involving 28,888 US patents and 9,196 inventors in the field of energy storage technology. This study contributes to advancing the theoretical understanding of technological knowledge ecosystems while also serving as a supplementary tool to explore organizational collaboration opportunities.}, archive = {J_ESWA}, author = {Jaemin Chung and Jaewoong Choi and Janghyeok Yoon}, doi = {10.1016/j.eswa.2025.129724}, journal = {Expert Systems with Applications}, month = {3}, pages = {129724}, shortjournal = {Expert Syst. Appl.}, title = {Exploring intra- and inter-organizational collaboration opportunities across a technological knowledge ecosystem: A metapath2vec approach}, volume = {298}, year = {2026}, }
• (2026). An improved actor-critic architecture with PPO for the traveling salesman problem. ESWA, 298, 129723. (www)

  @article{ , abstract = {The traveling salesman problem (TSP) is a classic NP-hard problem in combinatorial optimization with extensive practical applications. In this paper, we present an improved Actor-Critic architecture incorporating Proximal Policy Optimization (PPO) to effectively solve TSP. We introduce adaptive temperature scheduling, comprehensive state representation, and layer normalization to enhance learning stability. Experimental results demonstrate our Improved Actor-Critic approach achieves significant improvements ranging from 8.7 % to 55.9 % for different problem sizes compared to established reinforcement learning baselines including Q-Learning, SARSA, Double Q-Learning, Actor-Critic with Experience Replay (ACER), and Trust Region Policy Optimization (TRPO), with particularly strong performance on smaller instances between 20 to 100 cities. When testing on standard TSPLIB benchmarks, our method shows consistent advantages of 12 % to 33 % compared to classical approaches While tabular methods become computationally infeasible beyond 250 cities due to memory constraints, our approach maintains high solution quality for problems up to 1432 cities on our experimental setup (Intel® Core™i9-10900X CPU @ 3.70GHz × 20 with four NVIDIA Quadro RTX 5000 GPUs). Our ablation studies confirm the importance of each component in our proposed architecture, in which the improved state representation provides the most significant contribution to our model performance. This research significantly advances reinforcement learning approaches to combinatorial optimization, with practical implications for logistics, telecommunications, and manufacturing. The developed source code is available at: https://github.com/LetuQingge/TSP_Environment .}, archive = {J_ESWA}, author = {Hailemicael Lulseged Yimer and Pei Yang and Letu Qingge}, doi = {10.1016/j.eswa.2025.129723}, journal = {Expert Systems with Applications}, month = {3}, pages = {129723}, shortjournal = {Expert Syst. Appl.}, title = {An improved actor-critic architecture with PPO for the traveling salesman problem}, volume = {298}, year = {2026}, }
• (2026). Synchronization of semi-markov jump two-time-scale fuzzy neural networks dealing with dual-scale hybrid attacks and its application. ESWA, 298, 129718. (www)

  @article{ , abstract = {In this paper, the synchronization issue of the semi-Markov jump two-time-scale fuzzy neural networks dealing with dual-scale hybrid attacks is addressed, in which the dual-scale hybrid attacks mean that hybrid attacks can be encountered in different time scales transmission channels. Based on mismatched membership functions, a new ϵ -dependent combined synchronization controller is proposed to ensure the synchronization of the master semi-Markov jump two-time-scale fuzzy neural networks and the slave ones while it is capable of resisting independent dual-scale hybrid attacks. Through the development of a Lyapunov function incorporating the singular perturbation parameter ϵ , stability conditions and a computational approach for determining the synchronization controller gain are derived for semi-Markov jump two-time-scale fuzzy neural networks. Finally, some simulations and two encryption and decryption processes are used to show the effectiveness of obtained results.}, archive = {J_ESWA}, author = {Feng Li and Ya-Nan Wang and Lei Su and Sangmoon Lee}, doi = {10.1016/j.eswa.2025.129718}, journal = {Expert Systems with Applications}, month = {3}, pages = {129718}, shortjournal = {Expert Syst. Appl.}, title = {Synchronization of semi-markov jump two-time-scale fuzzy neural networks dealing with dual-scale hybrid attacks and its application}, volume = {298}, year = {2026}, }
• (2026). Automated recognition and measurement of cellular and dendritic microstructures in alloys via machine learning. ESWA, 298, 129717. (www)

  @article{ , abstract = {The dendritic or cellular morphologies of alloys and metals formed during casting processes significantly influence key properties such as mechanical strength, toughness, hardness, and electrical or thermal conductivities. The literature presents correlations between these properties and the microstructural length scale, which requires accurate characterization at the micrometer level. However, traditional evaluation methods demand extensive experimental efforts, including careful metallographic preparation, high-quality imaging, and a large number of measurements for statistical reliability - often relying on analyst proficiency. This study proposes a machine learning-based workflow tailored for the automated processing of microstructure images. The approach enables the autonomous measurement of key microstructural features while minimizing bias and inconsistencies among analysts. By integrating advanced image processing techniques with object detection algorithms based on Convolutional Neural Networks (CNNs), the method autonomously identifies microstructural morphologies and quantifies their spacing scales. Three model types-Cell, Dendrite, and Hybrid (exhibiting both dendritic and cellular features)-were trained and validated on using a dataset of 200 images. Among them, the Cell detection model achieved the highest performance, with a mean Average Precision (mAP) of 78.77 %, followed by the Hybrid (75.63 %) and Dendrite (72.87 %) models. Finally, the automated measurements models were applied to literature images and compared to reported microstructural growth correlations.}, archive = {J_ESWA}, author = {Guilherme Marim da Silva and Rafael Kakitani and Carlos Henrique da Silva Santos and Amauri Garcia and Noé Cheung}, doi = {10.1016/j.eswa.2025.129717}, journal = {Expert Systems with Applications}, month = {3}, pages = {129717}, shortjournal = {Expert Syst. Appl.}, title = {Automated recognition and measurement of cellular and dendritic microstructures in alloys via machine learning}, volume = {298}, year = {2026}, }
• (2026). Online reinforcement learning strategies driven artificial bee colony algorithm for bi-objective distributed reentrant flowshops scheduling problem with sequence-sependent setup time. ESWA, 298, 129716. (www)

  @article{ , abstract = {With the increasing complexity of manufacturing systems, reentrancy has become prevalent in many production environments. This study investigates a bi-objective distributed reentrant flow shop scheduling problem with sequence-dependent setup times (DRFSP-SDST). The objectives are to minimize the total energy consumption (TEC) and the maximum completion time (makespan), simultaneously. First, a bi-objective mathematical model for the DRFSP-SDST is formulated based on practical reentrant production scenarios. Second, the artificial bee colony (ABC) algorithm and its variants are employed to solve the DRFSP-SDST. According to the characteristics of the DRFSP-SDST, six local search operators are specifically designed to enhance the performance of the proposed algorithms. For promoting greener and more energy-efficient production, two speed-scaling strategies are developed. Third, two reinforcement learning (RL) algorithms, Q-learning and State-Action-Reward-State-Action (SARSA), are integrated into the iterative process as online learning strategies to guide the selection of high-quality local search strategies during the iterations of the proposed algorithms. For each RL algorithm, two distinct selection strategies for local search operators are designed. Finally, the effectiveness of the proposed enhancement strategies is evaluated through comprehensive numerical experiments on 36 benchmark instances. The performance of the proposed algorithms is further validated via the Friedman test. The experimental results and analysis demonstrate that the ABC algorithm enhanced by SARSA-based local search exhibits superior competitiveness in solving the DRFSP-SDST.}, archive = {J_ESWA}, author = {Ao Yao and Kaizhou Gao and Ponnuthurai Nagaratnam Suganthan and Hongyan Sang}, doi = {10.1016/j.eswa.2025.129716}, journal = {Expert Systems with Applications}, month = {3}, pages = {129716}, shortjournal = {Expert Syst. Appl.}, title = {Online reinforcement learning strategies driven artificial bee colony algorithm for bi-objective distributed reentrant flowshops scheduling problem with sequence-sependent setup time}, volume = {298}, year = {2026}, }
• (2026). Defects inspection system for building facades using drones and deep learning method. ESWA, 298, 129715. (www)

  @article{ , abstract = {Regular inspection and maintenance of building facades are essential for preserving structural integrity and aesthetic quality, especially in aging urban high-rises. While drone-based visual inspection powered by artificial intelligence (AI) offers benefits in speed, safety, and scalability, existing methods are typically limited to single defect types or uniform facade categories due to the challenges of detecting multi-scale defects in complex, heterogeneous environments. This study introduces an automated multiclass defects inspection system for building facades by integrating drone technology, an AI-driven segmentation platform, and automatic report generation. Central to the system is a segmentation AI model capable of detecting multiclass defects with orders-of-magnitude differences in scale across diverse facade backgrounds. To handle the pixel imbalance of defects ranging from fine cracks to large spalling and glass breakage, the model is built upon EfficientUNet++, trained on a carefully curated dataset and optimized using adjustable batch sizes and active learning rates to improve multi-scale feature learning and mitigate overfitting. Evaluations on validation and out-of-sample datasets demonstrate that the proposed model achieves superior performance across all defect classes. Real-world drone experiments further confirm the model’s practical applicability, with high recall rates in detecting spalling, water seepage, cracks, and glass breakage across different types of facades. This work pioneers a robust, scalable, and efficient AI-based framework for automated multiclass facade defect inspection, providing actionable information for engineers and supporting urban infrastructure maintenance.}, archive = {J_ESWA}, author = {Xiaoling Zhou and Robert Lee Kong Tiong}, doi = {10.1016/j.eswa.2025.129715}, journal = {Expert Systems with Applications}, month = {3}, pages = {129715}, shortjournal = {Expert Syst. Appl.}, title = {Defects inspection system for building facades using drones and deep learning method}, volume = {298}, year = {2026}, }
• (2026). Cross-modal multitask learning for automated quantitative characterization of infrastructure airhole defects. ESWA, 298, 129713. (www)

  @article{ , abstract = {Quantitative characterization of apparent quality defects in infrastructure is a crucial component of operations and maintenance. It enables rapid assessment of defect severity and supports the timely formulation of preventive strategies. However, a singular visual modality struggles to simultaneously ensure the dual tasks of defect detection and measurement accuracy. To solve these problems, this paper proposes a novel framework for cross-modal multitask learning networks, which comprehensively integrates the advantages of image detection and point cloud measurement. The pixel points identified in the image are mapped to their corresponding three-dimensional coordinates in the point cloud through intensive feature matching. A measurement strategy for the inherent characteristics of the defect is subsequently proposed. Based on prior knowledge of the defect, the area and volume of defects are quantified accurately. Finally, extensive experiments on detection, matching and measurement demonstrate the efficacy of the proposed method. The results provide a valuable reference for the quantitative characterization of infrastructure defects.}, archive = {J_ESWA}, author = {Yu Wang and Yingchao Dai and Xiaodong Gan and Zhengtao Yang and Zhou Wu}, doi = {10.1016/j.eswa.2025.129713}, journal = {Expert Systems with Applications}, month = {3}, pages = {129713}, shortjournal = {Expert Syst. Appl.}, title = {Cross-modal multitask learning for automated quantitative characterization of infrastructure airhole defects}, volume = {298}, year = {2026}, }
• (2026). Language proficiency assessment of autistic children using large language models. ESWA, 298, 129712. (www)

  @article{ , abstract = {Language impairment is a common comorbidity in children with autism spectrum disorder (ASD), and language proficiency assessment is a primary method for identifying such impairments. However, traditional assessment tools are often subjective and inefficient, while existing computer-assisted methods are limited by a narrow focus and insufficient use of natural language samples. To address these issues, this study proposes a framework for assessing children’s language abilities based on large language models (LLMs). We first preprocess the natural language samples from children and design multiple assessment dimensions and workflows. To enhance the stability of the assessment, we introduce a multi-expert voting mechanism and perform a comparative analysis of various large language models’ performance. The experimental results demonstrate a strong correlation between the framework’s assessment results and the Mullen Scales of Early Learning (MSEL) verbal developmental quotients, with a Pearson correlation coefficient of 0.8 ( p < 0.001). Furthermore, the results show that the multi-dimensional evaluation can accurately differentiate between ASD and typically developing (TD) children, achieving a classification accuracy of 0.98. These findings suggest that the proposed framework has significant potential for improving the accuracy of ASD identification.}, archive = {J_ESWA}, author = {Saige Qin and Min Liu and Tongquan Wei and Qiaoyun Liu}, doi = {10.1016/j.eswa.2025.129712}, journal = {Expert Systems with Applications}, month = {3}, pages = {129712}, shortjournal = {Expert Syst. Appl.}, title = {Language proficiency assessment of autistic children using large language models}, volume = {298}, year = {2026}, }
• (2026). EQUINAS: Equilibrium-guided differentiable neural architecture search. ESWA, 298, 129711. (www)

  @article{ , abstract = {Recent research has significantly mitigated the performance collapse issue in Differentiable Architecture Search (DARTS) by either refining architecture parameters to better reflect the true strengths of operations or developing alternative metrics for evaluating operation significance. However, the actual role and impact of architecture parameters remain insufficiently explored, creating critical ambiguities in the search process. To address this gap, we conduct a rigorous theoretical analysis demonstrating that the change rate of architecture parameters reflects the sensitivity of the supernet’s validation loss in architecture space, thereby influencing the derived architecture’s performance by shaping supernet training dynamics. Building on these insights, we introduce the concept of a Stable Equilibrium State to capture the stability of the bi-level optimization process and propose the Equilibrium Influential ( E I ) metric to assess operation importance. By integrating these elements, we propose EQUINAS, a differentiable NAS approach that leverages the Stable Equilibrium State to identify the optimal state during the search process and derives the final architecture using the E I metric. Extensive experiments across diverse datasets and search spaces demonstrate that EQUINAS achieves competitive test accuracy compared to state-of-the-art methods while significantly reducing search costs. Additionally, EQUINAS shows remarkable performance in Transformer-based architectures and excels in real-world applications such as image classification and text recognition.}, archive = {J_ESWA}, author = {Weisheng Xie and Xiangxiang Gao and Xuwei Fang and Hui Li and Chen Hang and Shaoyuan Li}, doi = {10.1016/j.eswa.2025.129711}, journal = {Expert Systems with Applications}, month = {3}, pages = {129711}, shortjournal = {Expert Syst. Appl.}, title = {EQUINAS: Equilibrium-guided differentiable neural architecture search}, volume = {298}, year = {2026}, }
• (2026). Freq-DETR: Frequency-aware transformer for real-time small object detection in unmanned aerial vehicle imagery. ESWA, 298, 129710. (www)

  @article{ , abstract = {Recent advancements in unmanned aerial vehicle (UAV) and remote sensing technologies have propelled UAV object detection to the forefront of computer vision research. Despite significant progress in deep learning-based detection algorithms, critical challenges persist in small object detection, including high-frequency information loss, inadequate multiscale feature representation, etc. To address these limitations, this paper proposes Freq-DETR, a frequency-aware real-time transformer detection framework leveraging frequency domain analysis to enhance edge detail preservation and global contextual modeling through three novel innovations. First, the frequency-enhanced convolution module (FECM) synergistically integrates spatial and frequency features via dual-branch processing; Second, the decoupled intra-feature scale interaction module (DSC-Clo block) facilitates the integration of high-frequency local and low-frequency global information; Finally, the attention-guided selective feature pyramid network (AGS-FPN) employs context-aware attention for high-level screening feature fusion. Extensive evaluations on the VisDrone2019 benchmark demonstrate that Freq-DETR outperforms the baseline RT-DETR by 4.9 % m a p @ 50 gain while maintaining computational efficiency. There are also remarkable improvements on both UAVDT and HIT-UAV datasets. Ablation investigations and visual interpretability analyses further confirm the complementary benefits of its frequency-domain components and the framework’s robustness in complex aerial scenarios.}, archive = {J_ESWA}, author = {Jiayi Chen and Ningzhong Liu and Han Sun and Yu Wang}, doi = {10.1016/j.eswa.2025.129710}, journal = {Expert Systems with Applications}, month = {3}, pages = {129710}, shortjournal = {Expert Syst. Appl.}, title = {Freq-DETR: Frequency-aware transformer for real-time small object detection in unmanned aerial vehicle imagery}, volume = {298}, year = {2026}, }
• (2026). Companion learning networks: A deep reinforcement learning algorithm with partner networks. ESWA, 298, 129709. (www)

  @article{ , abstract = {Existing deep reinforcement learning (DRL) agents suffer from severe reward instability during late-stage exploration, particularly when encountering novel states in complex continuous environments. A variety of existing studies focus on improving an agent’s reward exploration. However, they ignore the instability problem that arises when the agent faces new states in the later stages of exploration. This paper proposes a novel companion learning network (CLN) based on the idea that the guidance can accelerate human learning efficiency and reduce the risk of making mistakes. The CLN integrates a short-term partner network to intensively learn localized environmental patterns, offering adaptive action guidance for recent states. Simultaneously, a global Q-network dynamically incorporates the partner’s decaying guidance signals, balancing autonomous exploration with error mitigation. As training progresses, the partner’s influence gradually diminishes, allowing the Q-network to solidify robust policies without persistent dependence. Extensive experiments on four OpenAI Gym environments demonstrate that the CLN can significantly improve the exploration stability in most tested scenarios, achieving up to 49% reduction in late-stage reward standard deviation compared to baseline DRL methods.}, archive = {J_ESWA}, author = {Jin Xu and Jinfeng Bu and Yu Zhang and Jia-Dong Zhang and Chi-Yin Chow}, doi = {10.1016/j.eswa.2025.129709}, journal = {Expert Systems with Applications}, month = {3}, pages = {129709}, shortjournal = {Expert Syst. Appl.}, title = {Companion learning networks: A deep reinforcement learning algorithm with partner networks}, volume = {298}, year = {2026}, }
• (2026). Expensive multiobjective immune algorithm using a novel differential evolution in objective space. ESWA, 298, 129708. (www)

  @article{ , abstract = {Generating offspring solutions with strong convergence and diversity is critical when solving expensive multiobjective optimization problems due to the limited number of objective function evaluations. However, existing algorithms produce offspring solutions in the decision space, causing significant uncertainty in obtaining offspring with strong convergence and diversity. To address this issue, we devise a novel differential evolution based on the objective space rather than the decision space, called Differential Objective Evolution (DOE). Specifically, DOE generates objective values with strong convergence and diversity and then maps these values to the decision space to achieve high-quality offspring. Furthermore, we utilize a multiobjective immune algorithm to produce high-quality samples for effectively training the mapping in DOE. When compared with eleven recently proposed algorithms on 105 expensive multiobjective optimization problems, the experiments demonstrate the superiority of our algorithm and the contributions of DOE in both population convergence and diversity.}, archive = {J_ESWA}, author = {Yuchao Su and Wu Lin and Daxin Zhu and Anhui Tan and Ka-Chun Wong and Qiuzhen Lin}, doi = {10.1016/j.eswa.2025.129708}, journal = {Expert Systems with Applications}, month = {3}, pages = {129708}, shortjournal = {Expert Syst. Appl.}, title = {Expensive multiobjective immune algorithm using a novel differential evolution in objective space}, volume = {298}, year = {2026}, }
• (2026). Talk with graph: A fuzzy sentiment-aware framework for explainable recommendation. ESWA, 298, 129707. (www)

  @article{ , abstract = {User reviews reflect user preferences and item characteristics, optimizing the predictive accuracy and explanation generation of personalized recommendation systems. However, existing models face challenges due to subjective uncertainty in user feedback and a lack of transparency. Reviews often contain ambiguous emotional expressions, with the same product receiving positive, neutral, and negative sentiments. Many recommendation models assume alignment between ratings and review sentiments, but in practice, users may give high ratings while expressing dissatisfaction or vice versa. These inconsistencies complicate accurate modeling of user preferences. To address these issues, a Large Language Model (LLM)-driven sentiment-enhanced heterogeneous graph neural network framework is proposed. This framework jointly models interaction data and fuzzy sentiment information from reviews to improve both recommendation accuracy and explainability. By leveraging LLM with dual-prompt strategies, high-quality sentiment distributions and semantic insights are extracted. Review sentiments are then quantified using intuitionistic fuzzy numbers to address data sparsity and uncertainty, capturing implicit relationships between users, items, and entities in a sentiment-enhanced heterogeneous relational graph. A fuzzy sentiment-weighted graph convolutional network (FSGCN) is introduced for dynamic higher-order feature learning, adjusting sentiment weights based on user-item interactions and emotional context. The framework also integrates LLM-driven query interpretation to generate recommendations with transparent, context-aware rationales. This approach enables users to understand the reasoning behind recommendations, significantly enhancing explainability and trust.}, archive = {J_ESWA}, author = {Zhinan Li and Zhenyu Liu and Guodong Sa and Mingjie Hou and Jiacheng Sun and Jianrong Tan}, doi = {10.1016/j.eswa.2025.129707}, journal = {Expert Systems with Applications}, month = {3}, pages = {129707}, shortjournal = {Expert Syst. Appl.}, title = {Talk with graph: A fuzzy sentiment-aware framework for explainable recommendation}, volume = {298}, year = {2026}, }
• (2026). Geodesic-based path planning for port transfer robots on riemannian manifolds. ESWA, 298, 129706. (www)

  @article{ , abstract = {With the rapid intelligent transformation of the automotive industry and the surge in production volume, intelligent autonomous robots equipped with integrated perception and planning systems are playing an increasingly vital role in vehicle transfer operations. Optimizing dispatch paths of robots is essential for improving overall operational efficiency, yet achieving a balance among path length, feasibility, and safety margin remains a significant challenge. To address this issue, we propose a geodesic-based path planning method formulated on Riemannian manifolds. The approach jointly considers directional motion constraints, steering effort, and obstacle accessibility boundaries to construct a Riemannian metric tensor that encodes local path cost structures. This transforms the planning task into a geodesic shortest path problem, which is efficiently solved using the Geometric heat flow (GHF) method. The resulting paths naturally comply with kinematic constraints and exhibit strong obstacle-avoidance capabilities, significantly enhancing safety and executability. Extensive simulations and real-world experiments in high-density port yard environments demonstrate the practicality and robustness of the proposed method under complex spatial constraints and obstacle configurations.}, archive = {J_ESWA}, author = {Runjiao Bao and Junzheng Wang and Shoukun Wang}, doi = {10.1016/j.eswa.2025.129706}, journal = {Expert Systems with Applications}, month = {3}, pages = {129706}, shortjournal = {Expert Syst. Appl.}, title = {Geodesic-based path planning for port transfer robots on riemannian manifolds}, volume = {298}, year = {2026}, }
• (2026). A social balance theory-based modeling framework for group-to-empirical decision-making transition with cognitive inertia and trust propagation. ESWA, 298, 129705. (www)

  @article{ , abstract = {Inspired by empirical decision-making (EDM) processes, we propose a novel modeling framework where agents iteratively integrate social neighbors’ opinions into their cognitive inertia sequences (CISs), gradually prioritizing their accumulated CISs over time. This framework simulates the transition from group decision-making (GDM) to EDM through dynamic trust/distrust propagation and aggregation mechanisms grounded in social balance theory–capturing relational scenarios such as “a friend of a friend is a friend”, “a friend of an enemy is an enemy”, “an enemy of a friend is an enemy”, and “an enemy of an enemy is a stranger”. The paradigm incorporates two core mechanisms: (1) an endogenous cognitive inertia mechanism that uses the psychological serial-positioning effect to model cognitive inertia weights, accounting for primacy, recency, and U-shaped memory effects; and (2) an exogenous mechanism that quantifies comprehensive trust/distrust degrees via opinion similarity, stability similarity, and network structure similarity. To prevent followers from falling into cognitive freezing, a cluster leader-based consensus-reaching strategy is introduced. Extensive comparative experiments on real-world network datasets confirm the model’s effectiveness and robustness.}, archive = {J_ESWA}, author = {Jianglin Dong and Yiyi Zhao and Shangqun Mu and Haixia Mao and Jiangping Hu}, doi = {10.1016/j.eswa.2025.129705}, journal = {Expert Systems with Applications}, month = {3}, pages = {129705}, shortjournal = {Expert Syst. Appl.}, title = {A social balance theory-based modeling framework for group-to-empirical decision-making transition with cognitive inertia and trust propagation}, volume = {298}, year = {2026}, }
• (2026). OTPS-VO: Enhanced RGB-D odometry for indoor service robots leveraging structural features. ESWA, 298, 129704. (www)

  @article{ , abstract = {Simultaneous Localization and Mapping (SLAM) is essential for indoor service robots to achieve reliable navigation and mapping. While point and line features have been extensively utilized to enhance the accuracy of visual odometry (VO), current methods often overlook the rich geometric information embedded in the spatial relationships among structural lines. In particular, the parallelism and collinearity within groups of line segments are underexploited, and geometric constraints are typically applied only heuristically or post hoc, limiting robustness in low-texture and repetitive environments. To address these challenges, a robust VO system is proposed that integrates structural feature grouping with adaptive MW tracking. A unified feature extraction strategy is introduced to detect point and line features simultaneously, improving computational efficiency. To mitigate pose drift caused by unreliable line segments, a set of parallel line features is constructed based on local geometric constraints, and a novel reprojection error model is formulated to enhance pose estimation. Furthermore, a tracking strategy based on local Manhattan World (MW) structure is developed to ensure low-drift estimation across various structured indoor scenes. Extensive experiments on multiple public datasets and a custom-built service robot platform demonstrate that the proposed method outperforms existing state-of-the-art approaches under dynamic lighting conditions and in environments rich in lines and planes. The system also operates at a real-time speed of 30 frames per second, meeting the requirements of practical robotic applications.}, archive = {J_ESWA}, author = {Zhiyu Wang and Weili Ding and Ying Zhang and Changchun Hua}, doi = {10.1016/j.eswa.2025.129704}, journal = {Expert Systems with Applications}, month = {3}, pages = {129704}, shortjournal = {Expert Syst. Appl.}, title = {OTPS-VO: Enhanced RGB-D odometry for indoor service robots leveraging structural features}, volume = {298}, year = {2026}, }
• (2026). Adaptive gated meta graph retention network: A model for urban traffic flow prediction. ESWA, 298, 129703. (www)

  @article{ , abstract = {Traffic flow prediction is crucial for urban transportation systems. Existing models are still deficient in training efficiency and modeling dynamic spatial-temporal dependencies, various external factors, time-varying topology. Based on this, this paper introduces the Adaptive Gated Meta Graph Retention Network (AGMGRN), a novel model for spatial-temporal traffic flow prediction. Specifically, the AGMGRN integrates the attention mechanism with the retention network to model spatial-temporal dependencies. The AGMGRN proposes a gated dynamic connection block to enhance the model’s dynamic modeling capabilities. The AGMGRN considers the influence of external factors on traffic conditions through meta-learning approaches. The AGMGRN proposes an adaptive graph block to construct time-varying topologies. Experiments on four actual large-scale datasets demonstrate that the AGMGRN achieves superior prediction accuracy and high applicability.}, archive = {J_ESWA}, author = {Xing Li and Yuequan Bao}, doi = {10.1016/j.eswa.2025.129703}, journal = {Expert Systems with Applications}, month = {3}, pages = {129703}, shortjournal = {Expert Syst. Appl.}, title = {Adaptive gated meta graph retention network: A model for urban traffic flow prediction}, volume = {298}, year = {2026}, }
• (2026). DASF-GRL: Dynamic agent-scaling framework with game-augmented reinforcement learning for defensive counter-air operations. ESWA, 298, 129702. (www)

  @article{ , abstract = {Defensive Counter-Air (DCA) operations are pivotal for modern air defense, but existing studies are limited by static defender populations and oversimplified attacker models. We address these limitations with the Dynamic Agent-Scaling Framework with Game-Augmented Reinforcement Learning (DASF-GRL), which dynamically scales defender populations based on real-time threat levels. Specifically, we introduce a hybrid imitation-reinforcement training strategy that integrates attention mechanisms into critic networks to enable dynamic agent scaling. By incorporating a safety barrier function rooted in differential game theory, we constrain agents’ action spaces and enhance policy reliability. Furthermore, we developed a DCA simulation platform supporting reinforcement learning validation and designed a novel Apollonius-based penetration strategy for attackers to improve algorithmic robustness. Experiments demonstrate that DASF-GRL adaptively adjusts defender populations across scenarios involving 20, 40, and 60 attackers, markedly outperforming baseline methods in convergence speed and defense success rates. This framework offers novel theoretical paradigms and practical tools for intelligent decision-making in DCA environments.}, archive = {J_ESWA}, author = {Yuxuan Chen and He Luo and Guoqiang Wang}, doi = {10.1016/j.eswa.2025.129702}, journal = {Expert Systems with Applications}, month = {3}, pages = {129702}, shortjournal = {Expert Syst. Appl.}, title = {DASF-GRL: Dynamic agent-scaling framework with game-augmented reinforcement learning for defensive counter-air operations}, volume = {298}, year = {2026}, }
• (2026). Coordinated low-carbon economic scheduling of integrated electricity-heat-gas-hydrogen-methanol multi-microgrids considering electricity-methanol transaction. ESWA, 298, 129701. (www)

  @article{ , abstract = {The uneven spatial and temporal distribution of renewable energy resources poses significant challenges for multi-microgrid (MG) systems, resulting in high operational costs and low renewable energy utilization. To overcome these challenges, this work investigates a peer-to-peer electricity transaction and hydrogen-methanol-hydrogen technology-based methanol transaction among multi-MG. Besides, to realize net-zero emissions and carbon cycle utilization, the carbon capture system and hydrogen blending system are introduced into MG to reduce carbon dioxide emissions and capture and reform carbon dioxide for methanol synthesis equipment. Additionally, a cooperative operation model based on the Nash bargaining theory for multi-MGs under the transaction amount and price constraints of electricity and methanol is constructed. Due to the characteristics of non-convex and non-linear, the Nash bargaining is transformed into minimizing operation costs (sub-problem one) and maximizing payment benefits (sub-problem two). During the process of benefit allocation in sub-problem two, this work adopts a nonlinear energy sharing mapping method to quantify the comprehensive contribution rate of each MG to the multi-MG system, thereby achieving fair allocation of benefits. Finally, the alternating direction multiplier method is used to solve the model, effectively protecting the privacy of each MG. The simulation results demonstrate that a multi-MG system considering electricity and methanol transactions can effectively decrease carbon emissions and the total operational costs by 21.53% and 27.01% compared to only considering electricity transactions, respectively. Overall, the proposed electricity and methanol transactions strategy simultaneously reduces the overall system operation costs and carbon emissions, underscoring its advantages and significance.}, archive = {J_ESWA}, author = {Jiale Li and Bo Yang and Yiming Zhou and Hongchun Shu and Hongbiao Li and Dengke Gao and Lin Jiang}, doi = {10.1016/j.eswa.2025.129701}, journal = {Expert Systems with Applications}, month = {3}, pages = {129701}, shortjournal = {Expert Syst. Appl.}, title = {Coordinated low-carbon economic scheduling of integrated electricity-heat-gas-hydrogen-methanol multi-microgrids considering electricity-methanol transaction}, volume = {298}, year = {2026}, }
• (2026). Heterogeneous cloud resource allocation: A case study on real-time transcoding in live streaming. ESWA, 298, 129700. (www)

  @article{ , abstract = {The explosion in popularity of crowdsourced live streaming (CLS) has led to a huge increase in demand for cloud resources to support real-time video transcoding. CLS transcoding is real-time, geographically distributed and computationally intensive. Therefore, transcoding service providers need to cost-effectively utilize diverse heterogeneous cloud resources, while guaranteeing quality of service standards to ensure a good streaming experience for the viewers. To support the above, we developed a novel proactive-reactive resource allocation framework that optimizes the overall cost of supporting the CLS transcoding service using heterogeneous edge and cloud computing resources. The offline proactive policy evaluator aims to provide a good and adaptable resource usage plan in advance, matching the predicted demand with the heterogeneous resources. The reactive execution module monitors the actual demand online and controls the resource usage to compensate for deviations from the offline prediction. Our experiments show that the proposed approach leads to a cost reduction of 42 % compared to the fixed usage ratio strategy based on expert knowledge.}, archive = {J_ESWA}, author = {Yinuo Li and Jin-Kao Hao and Kwong Meng Teo and Liwei Song}, doi = {10.1016/j.eswa.2025.129700}, journal = {Expert Systems with Applications}, month = {3}, pages = {129700}, shortjournal = {Expert Syst. Appl.}, title = {Heterogeneous cloud resource allocation: A case study on real-time transcoding in live streaming}, volume = {298}, year = {2026}, }
• (2026). Adaptive multi-step path planning for multi-robot in dynamic environments based on hybrid optimization approach. ESWA, 298, 129699. (www)

  @article{ , abstract = {The multi-robot path planning problem requires algorithms with high convergence speed and accuracy, as well as the completeness of the search probability for the optimal path. The integration of metaheuristic algorithms in path planning has proven to be remarkably efficient. This paper introduces a novel hybrid metaheuristic algorithm, Beluga Whale-Crayfish Optimization (BWCOA), for enhanced global optimization in path planning applications. While the Crayfish Optimization (COA) demonstrates superior convergence speed, its inherent probabilistic path completeness remains suboptimal. To address this limitation, we present three key innovations: a dynamic probability completion mechanism, adaptive convergence acceleration factors, and balanced exploration–exploitation trade-off parameters. The proposed BWCOA synergizes Beluga Whale Optimization (BWO)’s basin-hopping capability with COA’s swarm intelligence through parallel combined exploration strategies. To prove its powerfulness, a series of comparative analyses were conducted between BWCOA and other leading algorithms across two comprehensive test function suites. The numerical experiment results underscore the significant superiority of BWCOA over its counterparts. In the context of path planning simulations, BWCOA demonstrated notable improvements over COA within the same number of function evaluations, with average enhancement rates of 6.49 %, 7.42 %, 15.09 %, 76.42 %, and 0.73 % across five evaluation metrics. Similarly, when compared to BWO on the same set of indicators, BWCOA showed average improvement rates of 22.39 %, 27.71 %, 70.53 %, 260.86 %, and 41.22 %. Furthermore, the running time of BWCOA is comparable to that of similar algorithms.}, archive = {J_ESWA}, author = {Liguo Yao and Guanghui Li and Taihua Zhang and Abdelazim G. Hussien and Yao Lu}, doi = {10.1016/j.eswa.2025.129699}, journal = {Expert Systems with Applications}, month = {3}, pages = {129699}, shortjournal = {Expert Syst. Appl.}, title = {Adaptive multi-step path planning for multi-robot in dynamic environments based on hybrid optimization approach}, volume = {298}, year = {2026}, }
• (2026). A two-stage evolutionary algorithm based on hybrid penalty strategy and its application to multi-UAV path planning. ESWA, 298, 129698. (www)

  @article{ , abstract = {Constrained optimization problems with complex and dynamic constraints pose significant challenges for evolutionary algorithms, as the constraints reshape the solution space and create conflicts between feasibility maintenance and global exploration. To address this issue, this study proposes TSC-PSODE, a two-stage evolutionary algorithm based on a hybrid penalty strategy. The algorithm employs an external penalty in the early stage to preserve population diversity and enhance exploration, while an internal penalty in the later stage accelerates convergence toward high-quality feasible solutions. In addition, a cooperative strategy combining differential evolution operators strengthens robustness and helps the population escape local optima. Experimental evaluations on the CEC2017 benchmark suite (the IEEE Congress on Evolutionary Computation 2017 benchmark) and multi-Unmanned Aerial Vehicle path planning tasks demonstrate that TSC-PSODE consistently outperforms state-of-the-art algorithms. The results confirm that the proposed method not only provides an effective mechanism for constraint handling, but also achieves a favorable balance between exploration and exploitation by maintaining diversity and accelerating convergence. In practical terms, TSC-PSODE is capable of generating safe and feasible flight paths for multiple UAVs in complex environments, highlighting its adaptability and competitiveness for real-world applications.}, archive = {J_ESWA}, author = {Eryang Guo and Yuelin Gao and Chenyang Hu}, doi = {10.1016/j.eswa.2025.129698}, journal = {Expert Systems with Applications}, month = {3}, pages = {129698}, shortjournal = {Expert Syst. Appl.}, title = {A two-stage evolutionary algorithm based on hybrid penalty strategy and its application to multi-UAV path planning}, volume = {298}, year = {2026}, }
• (2026). Critical commentary on “Reptile search algorithm (RSA): A nature-inspired meta-heuristic optimizer”. ESWA, 298, 129697. (www)

  @article{ , archive = {J_ESWA}, author = {Ngaiming Kwok}, doi = {10.1016/j.eswa.2025.129697}, journal = {Expert Systems with Applications}, month = {3}, pages = {129697}, shortjournal = {Expert Syst. Appl.}, title = {Critical commentary on “Reptile search algorithm (RSA): A nature-inspired meta-heuristic optimizer”}, volume = {298}, year = {2026}, }
• (2026). 2PADMS: Two-stage prediction and data migration strategy based on hard disk failure time. ESWA, 298, 129696. (www)

  @article{ , abstract = {With the explosive growth of global data volume, the usage of hard disk drives (HDDs) is also increasing rapidly. Consequently, the number of failed disks is continuously rising, which can affect storage service quality and even lead to data loss when failures occur.In recent years, the active fault-tolerant technology, which collects hard disks’ Self Monitoring Analysis and Reporting Technology (SMART) data-set, predicts hard disk failure by machine learning model, and repairs near-failure disks’ data to health disks in advance, has become a common research hotspot in both academia and industry. Aiming at the existing problems such as interference characteristics, inaccurate failure time prediction, competition of system resources between data migration and front service, this paper researches the two-stage prediction model and data migration strategy based on hard disk failure time, including the two-stage hard disk information feature selection method, the two-stage prediction method of hard disk failure time, and the data migration elastic system resource allocation strategy. Feature selection is performed by combining embedding methods with visualization, and the importance of the selected features is evaluated using a random forest model. Based on the feature importance, further refinement is carried out to obtain the final feature set. Before predicting the failure time of hard drives, XGBoost is first used in a voting manner to identify drives predicted to be faulty. Then, a trained Bidirectional Long Short-Term Memory network (Bidirectional LSTM) enhanced with a self-attention mechanism is employed to predict the exact failure time.Experimental results show that on the Backblaze dataset, the model achieves a mean absolute error of 1.24 when predicting failure times. The recall rate for predicting failures within 7 days reaches 98.79 %, the error rate is 0.30 %, the F1 score is 99.24 %, and the precision is 99.69 %. The elastic system resource allocation strategy for data migration improves business IOPS by 47.19 % and reduces latency by 38.68 %.}, archive = {J_ESWA}, author = {Huiyuan Qiang and Yuequan Li and Hongzhang Yang and Ping Wang and Yaofeng Tu and Shang Yang}, doi = {10.1016/j.eswa.2025.129696}, journal = {Expert Systems with Applications}, month = {3}, pages = {129696}, shortjournal = {Expert Syst. Appl.}, title = {2PADMS: Two-stage prediction and data migration strategy based on hard disk failure time}, volume = {298}, year = {2026}, }
• (2026). EEG emotion recognition through a domain-adversarial multi-feature fusion network. ESWA, 298, 129694. (www)

  @article{ , abstract = {Accurate recognition of EEG signals linked to emotions is crucial for neuroscience and human-computer interaction. However, variability in EEG emotion recognition among individuals results in inconsistent feature distributions and limited generalization across subjects. To enhance the robustness of the model, we propose a deep learning approach integrating a domain adversarial migration network with an attention mechanism. Initially, a feature extractor with a hierarchical architecture (low-medium-high levels) is employed to capture multi-scale EEG features, which are then normalized for age and encoded for genderand education before being aligned with EEG features through spatio-temporal replication. Subsequently, global distribution alignment is achieved using multi-kernel maximum mean difference (MK-MMD), subdomain adversarial alignment is accomplished with a gradient inversion layer (GRL), and decision boundary clarity is enhanced through joint emotion classification loss. The generalization capability and effectiveness of the model are validated using the DEAP and DREAMER datasets, offering insights for cross-subject emotion recognition research and applications.}, archive = {J_ESWA}, author = {Weitong Sun and Yuping Su and Yumei Zhang and Xiaojun Wu}, doi = {10.1016/j.eswa.2025.129694}, journal = {Expert Systems with Applications}, month = {3}, pages = {129694}, shortjournal = {Expert Syst. Appl.}, title = {EEG emotion recognition through a domain-adversarial multi-feature fusion network}, volume = {298}, year = {2026}, }
• (2026). An acoustic-electromagnetic detection system based on joint parameter estimation of atomic norm algorithm. ESWA, 298, 129692. (www)

  @article{ , abstract = {The time delay and Doppler shift parameters in radar system echo signals serve as effective tools for multi-target identification and localization in covert environments. However, the nonlinear characteristics of stationary targets are often masked by surrounding environmental factors, and traditional joint parameter estimation algorithms tend to suffer from high computational complexity and errors during demodulation. To address these challenges, this paper proposes an acoustic-electromagnetic intermodulation detection system based on a novel atomic-paradigm algorithm, which ensures localization accuracy with minimal computational complexity and zero false alarms. Specifically, the system excites the target by introducing acoustic field energy coupling, generating discernible micromotion features. The resulting acoustically modulated signal is then modeled as a two-dimensional line spectral estimation problem, capturing the target’s time delay and Doppler shift. Furthermore, the joint parameter estimation algorithm is enhanced by relaxing dyadic constraints under sufficient conditions. In our experiments, a harmonic radar physical system is constructed to simultaneously localize and measure multiple non-clustered micromotion targets. The recognition accuracy is quantitatively evaluated using a classical neural network model, achieving 86.9 % accuracy across five classified targets. The improved algorithm’s performance in joint parameter estimation is also assessed under varying signal-to-noise ratios and demodulation error rates, with a detailed time complexity analysis provided.}, archive = {J_ESWA}, author = {Sheng Wu and Yilin Cai and Yijing Zheng and Dingzhao Li and Hongjun Lai and Haixin Sun}, doi = {10.1016/j.eswa.2025.129692}, journal = {Expert Systems with Applications}, month = {3}, pages = {129692}, shortjournal = {Expert Syst. Appl.}, title = {An acoustic-electromagnetic detection system based on joint parameter estimation of atomic norm algorithm}, volume = {298}, year = {2026}, }
• (2026). A novel convex-hull-based algorithm for classification problems with imbalanced and overlapping data. ESWA, 298, 129691. (www)

  @article{ , abstract = {Classification is a fundamental task in supervised machine learning. This problem becomes challenging when dealing with imbalanced and overlapping datasets. In such cases, learning algorithms often perform well in identifying the labels of majority class data points but exhibit high error rates in predicting the minority class. This paper proposes an innovative method based on the convex-hull concept to enhance classification for imbalanced and overlapping datasets. Unlike undersampling approaches that may lead to the loss of valuable information, our method focuses on preserving the data. The process begins by clustering the data points for each class separately in such a way that no points from the opposite class fall within the convex-hull of each cluster. Then, the support vector machine (SVM) is used to separate every cluster of a given class from the data points of the opposite class. Afterward, data points inside the SVM boundaries are considered as non-overlapping, while those outside the SVM boundaries are identified as overlapping data. The XGBoost algorithm is then employed to classify the data points within the overlapping region. Extensive experiments on a variety of simulated and real-world datasets confirm the effectiveness of the proposed method in terms of various evaluation metrics, compared to existing relevant algorithms for handling imbalanced and overlapping datasets.}, archive = {J_ESWA}, author = {Farnaz Hooshmand and Sogol Peik-Mortazavi}, doi = {10.1016/j.eswa.2025.129691}, journal = {Expert Systems with Applications}, month = {3}, pages = {129691}, shortjournal = {Expert Syst. Appl.}, title = {A novel convex-hull-based algorithm for classification problems with imbalanced and overlapping data}, volume = {298}, year = {2026}, }
• (2026). Unbiased representation learning via feature decoupling network for cross-scene hyperspectral image classification. ESWA, 298, 129690. (www)

  @article{ , abstract = {For hyperspectral cross-scene classification (HSIC) tasks, the model is trained on the source domains (SDs) and applied directly to the unseen target domains (TDs). For this domain generalization (DG) challenge, a significant issue is the contradiction between the overparameterized model and the limited training domain, which results in the absorption of spurious correlations from environmental features. To alleviate this contradiction, this study proposes a domain extension generator with a feature decoupling network (FDNet). The generator initially decouples the SD into reflectance and shading components, treating them as causal and environmental features, respectively. Considering possible causal residues in environmental features, the shading components are shuffled by style to eliminate undesired correlations. Then, the reflectance and sparsified shading are reconstructed for extension, enriching the training diversity without environmental interference. In addition, to enhance the stability of class-level causal representation, a supervised aggregation strategy is designed to minimize the intra-class distance of the reflectance domain, and supervised contrastive learning is employed to enhance the class-domain semantic consistency information. Comparative analysis with advanced domain generalization and adaptation approaches on three HSI datasets validates the superiority in accuracy and Kappa coefficient metrics of the proposed method.}, archive = {J_ESWA}, author = {Hanqing Zhao and Lianlei Lin and Zongwei Zhang and Sheng Gao and Junkai Wang}, doi = {10.1016/j.eswa.2025.129690}, journal = {Expert Systems with Applications}, month = {3}, pages = {129690}, shortjournal = {Expert Syst. Appl.}, title = {Unbiased representation learning via feature decoupling network for cross-scene hyperspectral image classification}, volume = {298}, year = {2026}, }
• (2026). Automated prompt engineering pipelines: Fine-tuning LLMs for enhanced response accuracy. ESWA, 298, 129689. (www)

  @article{ , abstract = {This paper presented an Automatic Role Prompting System that seeks to improve the performance of the Large Language Model (LLM) by allowing models to assume varied roles through role-based prompting and, as a result, qualitatively improve the relevance of outputs. Our Automatic Role Prompting System’s target audience is people who do not have domain knowledge. The guiding framework (consisting of an Automated Script for discovering roles and fields layered on top of prompt engineering, and Natural Language Inference (NLI) models trained in advance), was robustly tested through the use of three datasets: our set of 1990 curated prompts, WikiQA, and the AwesomeChatGPTPrompts. We implemented a novel evaluation strategy using GPT-Eval, which scales prompts according to completeness, clarity, and relevance. We found substantially better performance than traditional rule- and template-based approaches, yielding accuracy improvements as high as 97.6 %. Overall, this work demonstrates the promise of an Automated Role Prompting System to help people engage and work more effectively and efficiently with Large Language Models (LLMs).}, archive = {J_ESWA}, author = {Samar Hendawi and Tarek Kanan and Mohammed Elbes and Ala Mughaid and Shadi Alzu’bi}, doi = {10.1016/j.eswa.2025.129689}, journal = {Expert Systems with Applications}, month = {3}, pages = {129689}, shortjournal = {Expert Syst. Appl.}, title = {Automated prompt engineering pipelines: Fine-tuning LLMs for enhanced response accuracy}, volume = {298}, year = {2026}, }
• (2026). A hybrid fusion network using convolutional vision transformers for landslide identification. ESWA, 298, 129688. (www)

  @article{ , abstract = {Convolutional neural networks (CNN) have made great strides in the segmentation of remote sensing images, but they still have certain inherent drawbacks when working with small targets and complex geological structures. These drawbacks include incomplete contextual information integration, blurry edges, and inaccurate target localisation. This study suggests using a hybrid CNN-Transformer network to improve the segmentation of landslide regions in high-resolution remote sensing images in order to overcome these difficulties. A comprehensive investigation has been conducted on the use of CNN and transformers to accomplish the task of semantic segmentation. According to experimental data, our model outperforms the state-of-the-art CNN-based, Transformer-based, and even CNN-plus-Transformer combination models for image segmentation tasks by a large margin. When it comes to landslide area segmentation, it performs exceptionally well.}, archive = {J_ESWA}, author = {S. Sreelakshmi and S.~S. Vinod Chandra}, doi = {10.1016/j.eswa.2025.129688}, journal = {Expert Systems with Applications}, month = {3}, pages = {129688}, shortjournal = {Expert Syst. Appl.}, title = {A hybrid fusion network using convolutional vision transformers for landslide identification}, volume = {298}, year = {2026}, }
• (2026). Towards intelligent online cross-selling. ESWA, 298, 129686. (www)

  @article{ , abstract = {The ubiquity of online cross-selling for fashion demands a large number of qualified outfit compositions. This paper targets practical and intelligent online cross-selling by providing a more accurate fashion compatibility model and a reliable evaluation protocol for evaluating the fashion compatibility model. Firstly, a Hierarchical Outfit Network (HON) is proposed to leverage multi-layer relations among attributes, items, and outfits. The awareness of multiple relations hidden in various outfits enables the HON to outperform all state-of-the-art methods on fill-in-the-blank (FITB) accuracy and compatibility Area Under Curve (AUC) on the Maryland and Type-aware test sets. Meanwhile, a new evaluation protocol is introduced to assess the fashion compatibility model more objectively and accurately, namely, Aesthetic 100 (A100). A100 possess three desirable characteristics: 1) Completeness . All types of standards in the fashion aesthetic system are covered through two tests, namely LAT (Liberalism Aesthetic Test) and AAT (Academicism Aesthetic Test); 2. Reliability . It is an agnostic protocol and consistent with major indicators. It provides an objective and fair assessment for model comparison. 3. Explainability . A100 assesses the model on more fine-grained dimensions, e.g. , Color, Material, and Balance demonstrating its superiority in identifying essential characteristics of fashion aesthetics. Experimental results demonstrate the progress of A100 in the aspects of Reliability and Explainability. The evaluation results on A100 also show the generalization ability of HON from both quantitative and qualitative perspectives. Finally, solutions for multiple applications in fashion retailing are proposed to show how HON can be utilized to help online cross-selling.}, archive = {J_ESWA}, author = {Kaicheng Pang and Xingxing Zou and Zowie Broach and Waikeung Wong}, doi = {10.1016/j.eswa.2025.129686}, journal = {Expert Systems with Applications}, month = {3}, pages = {129686}, shortjournal = {Expert Syst. Appl.}, title = {Towards intelligent online cross-selling}, volume = {298}, year = {2026}, }
• (2026). GCMF-net: Global-aware cross-attention and mask-guided fusion for multispectral object detection. ESWA, 298, 129679. (www)

  @article{ , abstract = {Multispectral Object Detection has shown significant advantages under various lighting and weather conditions, with efficient fusion of RGB and thermal information playing a key role. Previous studies have demonstrated the effectiveness of feature fusion based on convolutional neural networks, but limited local feature interactions hinder the capture of global complementary information. To address these limitations, some methods adopt complex fusion strategies to enhance complementary feature extraction, yet fail to mitigate feature redundancy, which negatively impacts detection performance. In this paper, we propose a novel Global-aware Cross-attention and Mask-guided Fusion (GCMF) module for Multispectral Object Detection, following a fusion-refinement paradigm. In the fusion stage, we first introduce Efficient Channel Attention with Weighted Max-Pooling (ECA-WM) to focus on key information within each modality and achieve implicit alignment before fusion. Subsequently, the Global-aware Cross-Attention Transformer (GCAT) effectively captures complementary cross-modal information and models global features. In the refinement stage, the Mask-guided Refinement Strategy (MRS) generates segmentation masks to distinguish target features from the background. These masks are applied before and after cross-modal interaction to highlight target-relevant information while suppressing irrelevant features, resulting in highly discriminative fused representations. Extensive experimental comparisons demonstrate that the proposed GCMF fusion strategy achieves state-of-the-art performance on the publicly available FLIR, LLVIP and DVTOD datasets, with absolute improvements of 2.8 %, 4.2 % and 4.0 % over the previous methods, respectively. Moreover, the proposed strategy is general and effective, making it adaptable to various detection frameworks.}, archive = {J_ESWA}, author = {Zhong Qu and Jin Yang and Shufang Xia}, doi = {10.1016/j.eswa.2025.129679}, journal = {Expert Systems with Applications}, month = {3}, pages = {129679}, shortjournal = {Expert Syst. Appl.}, title = {GCMF-net: Global-aware cross-attention and mask-guided fusion for multispectral object detection}, volume = {298}, year = {2026}, }
• (2026). Oriented bounding box detection algorithm for dense scenarios of robotic arm operation. ESWA, 298, 129678. (www)

  @article{ , abstract = {In the context of the national promotion of intelligent manufacturing and the ’Industry4.0’ strategy, the demand for intelligent robotic arms in industrial production has steadily increased. However, challenges such as occlusion, significant object scale variations, and strict real-time requirements have made target detection in densely packed environments more challenging. This study, based on the YOLOv11 algorithm, proposes an efficient oriented bounding box detection method aimed at improving the model’s performance in feature extraction, computational efficiency, and network lightweighting to tackle target detection challenges in dense industrial settings. A Dynamic-Cross-Stage-Dual-Conv module was designed to enhance the Bottleneck section, employing a parallel dual-branch structure for local feature extraction and global context fusion. Simultaneously, a CIoU loss function with geometric perception was introduced to improve object localization accuracy and strengthen the network’s ability to handle densely stacked objects. Next, a Modulated-Deform-Conv deformable convolution module was integrated into the Backbone structure, dynamically adjusting the convolution kernel sampling positions, enabling the network to learn deformation features in dense scenes, improving adaptability to shape and scale variations while reducing computational load. Additionally, a C3K2_FasterBlock lightweight structure, utilizing partial convolutions and sparse connections, was proposed to minimize redundant calculations and optimize feature interactions. On a custom-built dense object dataset, the model achieved a 2.9 % increase in mAP@0.5 and reduced computational cost by 14 %. Finally, the improved model was deployed on the Jetson Orin Nano development board, demonstrating its practical value in robotic arm recognition and grasping tasks in dense industrial environments, offering a new paradigm for future applications.}, archive = {J_ESWA}, author = {Jinshun Dong and Lixia Deng and Dapeng Wan and Chenxu Liu and Jianqin Yin and Meiqi Guo and Hongyu Zhang and Shoujun Lin and Haiying Liu and Lida Liu}, doi = {10.1016/j.eswa.2025.129678}, journal = {Expert Systems with Applications}, month = {3}, pages = {129678}, shortjournal = {Expert Syst. Appl.}, title = {Oriented bounding box detection algorithm for dense scenarios of robotic arm operation}, volume = {298}, year = {2026}, }
• (2026). CrossModalNet: A dual-modal object detection network based on cross-modal fusion and channel interaction. ESWA, 298, 129677. (www)

  @article{ , abstract = {Object detection in complex environments, particularly under conditions such as low illumination or adverse weather, remains a significant challenge. Multispectral detection techniques that integrate visible and infrared imagery offer a promising solution by leveraging complementary modality information. However, substantial discrepancies between these modalities hinder traditional fusion strategies, which often fail to adaptively align and integrate features, resulting in information loss and diminished detection performance. To overcome this limitation, we propose CrossModalNet, a novel cross-modal fusion and channel interaction framework. CrossModalNet comprises two key modules: Convolutional Attention Interaction Module (CAIM) and Bidirectional Cross-Modal Attention Module (BiCAM). CAIM enables effective cross-modal integration across varying semantic levels by employing convolutional attention mechanisms combined with pixel-wise channel guidance. In parallel, BiCAM enhances inter-modal feature complementarity by modeling channel-level interactions through bidirectional attention pathways. Extensive experiments conducted on four public multispectral datasets, FLIR, LLVIP, M3FD, and VEDAI, demonstrate that the proposed method consistently outperforms existing state-of-the-art approaches across multiple performance metrics. Moreover, with an inference speed of 12.6 FPS on embedded platforms, the proposed model is suitable for real-time deployment.}, archive = {J_ESWA}, author = {Hanyun Li and Linsong Xiao and Lihua Cao and Di Wu and Yangfan Liu and Yi Li and Yunfeng Zhang and Haiyang Bao}, doi = {10.1016/j.eswa.2025.129677}, journal = {Expert Systems with Applications}, month = {3}, pages = {129677}, shortjournal = {Expert Syst. Appl.}, title = {CrossModalNet: A dual-modal object detection network based on cross-modal fusion and channel interaction}, volume = {298}, year = {2026}, }
• (2026). IPMMG: Information propagation with multi-granularity morphology-guided for nuclear segmentation and classification. ESWA, 298, 129676. (www)

  @article{ , abstract = {Nuclear segmentation and classification play a crucial role in pathological image analysis. However, it is frequently challenged by blurred nuclear boundaries and complex structures in digital pathology slides, due to factors such as staining techniques and imaging methods, posing a significant challenge for accurate segmentation and classification. To this end, we propose a novel and efficient approach for nuclear identification, termed Information Propagation with Multi-Granularity Morphology-Guided Network (IPMMG). Specifically, IPMMG progressively captures edge morphology information from different network layers while simultaneously incorporating structural morphology features at multiple granularities. By explicitly propagating features related to both the edge and the structure, our approach constrains semantic features to focus on contours of the region of interest in the nuclear segmentation task, thus mitigating the challenge of blurred morphology. Experiments on public datasets demonstrate that IPMMG achieves state-of-the-art (SOTA) performance in segmentation, as measured by Dice and IoU scores, while also attaining competitive results in classification with DQ, SQ, and PQ metrics. In particular, our proposal IPMMG excels in handling nuclei with blurred edges and complex structures.}, archive = {J_ESWA}, author = {Dawei Fan and Jun Li and Chengfei Cai and Lihui Lin and Riqing Chen and Yanping Chen and Lifang Wei}, doi = {10.1016/j.eswa.2025.129676}, journal = {Expert Systems with Applications}, month = {3}, pages = {129676}, shortjournal = {Expert Syst. Appl.}, title = {IPMMG: Information propagation with multi-granularity morphology-guided for nuclear segmentation and classification}, volume = {298}, year = {2026}, }
• (2026). MECA: Modular editing via customized expert networks and adaptors in large language models. ESWA, 298, 129675. (www)

  @article{ , abstract = {Updating language models with new information through targeted edits without resorting to expensive full model retraining remains a critical challenge, particularly when aiming to preserve pre-existing capabilities. In this work, we introduce M odular E diting via C ustomized expert networks and A daptors) (MECA), a unified framework that selectively integrates new knowledge into language models. MECA employs a module-level deferral router to evaluate whether incoming queries fall within the scope of existing edit requests. Queries are then dynamically routed to either customized editing experts or key-value adaptors. This modular strategy ensures that updates are localized, thereby mitigating risks of unintended alterations on unrelated outputs. We validate our approach on sequential editing tasks using Llama2-7B, Llama2-13B and Falcon 11B, benchmarked across two diverse datasets ZsRE and Hallucination. Experimental results show that MECA consistently outperforms several state-of-the-art knowledge editing techniques, achieving improved integration of new information while preserving the model’s original performance. Our analysis further demonstrates that the deferral routing mechanism for selecting modules effectively balances editing precision with overall model stability.}, archive = {J_ESWA}, author = {Roseline Nyange and Shanbao Qiao and Seung Hoon Na}, doi = {10.1016/j.eswa.2025.129675}, journal = {Expert Systems with Applications}, month = {3}, pages = {129675}, shortjournal = {Expert Syst. Appl.}, title = {MECA: Modular editing via customized expert networks and adaptors in large language models}, volume = {298}, year = {2026}, }
• (2026). An explainable hybrid bio-inspired feature selection framework using RSVP-evoked p300 EEG signals for identity authentication. ESWA, 298, 129674. (www)

  @article{ , abstract = {EEG-based personal identification confronts critical hurdles, including high dimensionality, noise, and real-time variability. While RSVP and P300 paradigms provide cognitive-response-driven security, feature extraction challenges prevent practical deployment. Although deep learning has addressed unstructured EEG data, pinpointing optimal RSVP and P300-specific features remains an unresolved issue. To overcome these limitations, we introduced a hybrid GWO-MSE-XAI framework integrating Grey Wolf Optimization (GWO), Multiscale Entropy (MSE), and SHAP-based Explainable AI (XAI) to select the most relevant features from RSVP-evoked P300 EEG signals. The framework prioritizes discriminative feature selection, improves class separability, and incorporates a hybrid cross-entropy loss function fused with Fisher’s score-based feature selection. Benchmark-driven optimization refines EEG-specific feature subsets, while evaluation using classifiers (Random Forest, LightGBM, CatBoost, XGBoost) demonstrates substantial dimensionality reduction, faster convergence, and superior performance (98.89% accuracy). Experimental results confirm robustness, scalability, and enhanced interpretability, positioning the framework as a viable solution for EEG-based identity authentication in real-world RSVP and P300 applications.}, archive = {J_ESWA}, author = {S Abinayaa and S.S. Sridhar}, doi = {10.1016/j.eswa.2025.129674}, journal = {Expert Systems with Applications}, month = {3}, pages = {129674}, shortjournal = {Expert Syst. Appl.}, title = {An explainable hybrid bio-inspired feature selection framework using RSVP-evoked p300 EEG signals for identity authentication}, volume = {298}, year = {2026}, }
• (2026). Interlink reconfiguration against cascading failures on cyber-physical power systems based on an improved memetic algorithm. ESWA, 298, 129673. (www)

  @article{ , abstract = {Cyber-physical power systems (CPPSs) are the deep integration of advanced information and other technologies applied to the grid to achieve fundamental changes in the power industry. Thus, it is very significant to enhance the robustness of the power communication system in order to ensure security. For instance, when an interdependent CPPSs is under attack, the failure can diffuse along interconnect topology to the whole system, even causing a system crash. Recall that most of the existing models are one-to-one coupling structures. Drawing close to reality, we develop a modified memetic algorithm to reconstruct CPPSs with multiple-to-multiple interlinks, called MA-Multiple, which is dedicated to improving the robustness of CPPSs. To improve the accuracy of the optimal solution of the MA-Multiple, we devise a crossover operator (CO) based on the null model to increase population diversity. Further improving the suitability of algorithms on CPPSs, we propose a new Kirchhoff centrality (KIC) based on the approximation of the inverse matrix to measure network connectivity and design a local search operator (LSO). In the experiment, we comprehensively compare the MA-Multiple with four closely relevant methods in different scenarios. The results imply that MA-Multiple performs better in enhancing network robustness and resisting attacks.}, archive = {J_ESWA}, author = {Wei Lin and Li Xu and Yuexin Zhang and Jie Li}, doi = {10.1016/j.eswa.2025.129673}, journal = {Expert Systems with Applications}, month = {3}, pages = {129673}, shortjournal = {Expert Syst. Appl.}, title = {Interlink reconfiguration against cascading failures on cyber-physical power systems based on an improved memetic algorithm}, volume = {298}, year = {2026}, }
• (2026). Multi-objective portfolio optimization for stock return prediction using machine learning. ESWA, 298, 129672. (www)

  @article{ , abstract = {This paper presents a novel approach that integrates stock return prediction with the mean–variance (MV) model to enhance the performance of the original model. Firstly, stock returns are predicted using machine learning algorithms, including Robust Linear Regression (OLS-H), Random Forest (RF), and Long Short-Term Memory Networks (LSTM), to select a pre-screened stock pool composed of stocks with high predicted returns. Secondly, a linear weighting method combines the predictions above with the MV model, constructing the Mean-Variance-Forecast Error (MVF) model and determining the investment proportions for the pre-selected stocks. Finally, empirical research is conducted using the components of the CSI 300 Index as sample data. The results indicate that the RF + MVF model outperforms other models and the CSI 300 Index in return and risk metrics. At the same time, a sensitivity analysis of relevant parameters further confirms that considering return uncertainty is beneficial for improving the out-of-sample performance of the MV model.}, archive = {J_ESWA}, author = {Meiyu Huang and Shili Dang and Miraj Ahmed Bhuiyan}, doi = {10.1016/j.eswa.2025.129672}, journal = {Expert Systems with Applications}, month = {3}, pages = {129672}, shortjournal = {Expert Syst. Appl.}, title = {Multi-objective portfolio optimization for stock return prediction using machine learning}, volume = {298}, year = {2026}, }
• (2026). EKT-ML: An efficient knowledge tracing model with multi-task learning. ESWA, 298, 129671. (www)

  @article{ , abstract = {Knowledge Tracing (KT) aims to trace a student’s mastery of knowledge, known as knowledge states, and has become a popular research area, with Self-Attention (SA)-based KT models achieving the state-of-the-art performance. However, existing SA-based KT models seem to still have issues that need further investigation. Firstly, there commonly exists incorrect question-knowledge concept (Q-KC) mapping, yet most models fail to address this issue. Secondly, existing SA-based KT models suffer from high time complexity due to their extensive use of the SA mechanism. Finally, from real-world datasets, we observe that there exists a repeated attempts pattern which is often overlooked by existing KT models. Motivated by the above observations, we propose a novel Efficient Knowledge Tracing Model with Multi-task Learning (EKT-ML), an SA-based model with three crucial features. Firstly, we formulate the KT as a Multi-task Learning, with Q-task and KC-task as two tasks; by training them simultaneously and treating Q-KC mapping as shared information, the proposed EKT-ML tends to mitigate the impact of incorrect Q-KC mapping. Moreover, we propose an SVD-MLP component to replace the initial SA layers commonly used in existing SA-based KT models, thereby reducing the time complexity of EKT-ML. Finally, experimental results show that EKT-ML improves performance by up to 8.84 % across four metrics on three widely used datasets. Furthermore, it demonstrates that the EKT-ML has reduced time complexity compared with the benchmark baseline.}, archive = {J_ESWA}, author = {Wei Liu and Bo Yang and Haotian Su and Yaowei Wang and Qing Li}, doi = {10.1016/j.eswa.2025.129671}, journal = {Expert Systems with Applications}, month = {3}, pages = {129671}, shortjournal = {Expert Syst. Appl.}, title = {EKT-ML: An efficient knowledge tracing model with multi-task learning}, volume = {298}, year = {2026}, }
• (2026). Curriculum learning for a hybrid approach for aspect-based sentiment analysis. ESWA, 298, 129669. (www)

  @article{ , abstract = {In the past years, the amount of unstructured online review data has grown exponentially. Many people express their opinions about different aspects of goods and services on the Web. Aspect-Based Sentiment Analysis (ABSA) automatically extracts the sentiments with respect to aspects given in a sentence. We improve the training procedure of the state-of-the-art Hybrid Approach for Aspect-Based Sentiment Analysis with deep contextual word embeddings and hierarchical attention (HAABSA++). In this method, a domain sentiment ontology is used as a main classifier, and if it is not conclusive, a neural network is employed as a back-up. We extend the training of the neural network by incrementally adding more difficult instances, also known as curriculum learning. Restaurant reviews obtained from the SemEval-2015 and SemEval-2016 datasets are used to evaluate the effect of implementing curriculum learning. Using baby steps curriculum learning and a specific curriculum strategy, the accuracy of HAABSA++ is improved from 86.3 % to 87.5 %.}, archive = {J_ESWA}, author = {Nana Lange and Flavius Frasincar and Maria Mihaela Truşcǎ}, doi = {10.1016/j.eswa.2025.129669}, journal = {Expert Systems with Applications}, month = {3}, pages = {129669}, shortjournal = {Expert Syst. Appl.}, title = {Curriculum learning for a hybrid approach for aspect-based sentiment analysis}, volume = {298}, year = {2026}, }
• (2026). A new edge detection method for noisy image based on discrete fractional wavelet transform and improved canny algorithm. ESWA, 298, 129668. (www)

  @article{ , abstract = {Edge detection is a key technique for extracting object structures and region boundaries in images, serving as an important foundation for visual tasks such as image segmentation and object recognition. However, during real-world image acquisition, various types of noise are inevitably introduced into the images. Traditional edge detection methods suffer significant performance degradation in noisy environments, often resulting in false edges or missing true edges. To address this issue, this paper proposes a novel edge detection method for noisy images. The method begins by adaptively selecting an optimal fractional order p , based on the distribution characteristics of the image’s subband modulus coefficients. This order is then used to perform a p -order discrete fractional wavelet transform (DFRWT) on the noisy image. Then, within the DFRWT domain, an enhanced Canny algorithm is applied to detect edges. This algorithm improves upon the standard method by replacing the traditional gradient operator with a more robust fractional-order Sobel operator to compute the gradient magnitude. This detection process is performed on both the low- and high-frequency subbands to capture features at different scales. Finally, the edge images from the low- and high-frequency components are reconstructed to obtain the final edge detection result. Experimental results demonstrate that, compared to four representative edge detection algorithms, the proposed method exhibits superior noise robustness and edge preservation capability in noisy environments.}, archive = {J_ESWA}, author = {Xiaozhong Yang and Chunmeng Li}, doi = {10.1016/j.eswa.2025.129668}, journal = {Expert Systems with Applications}, month = {3}, pages = {129668}, shortjournal = {Expert Syst. Appl.}, title = {A new edge detection method for noisy image based on discrete fractional wavelet transform and improved canny algorithm}, volume = {298}, year = {2026}, }
• (2026). Physics-constrained EBSD image inpainting via adversarial graph learning: Bridging crystallographic rules and multimodal deep learning. ESWA, 298, 129667. (www)

  @article{ , abstract = {Electron Backscatter Diffraction (EBSD) is a crucial characterisation method in materials engineering. The reliability of EBSD data is essential in the aerospace, nuclear, and automotive industries, as material performance greatly affects operational safety. While industrial practice makes perfect EBSD data difficult, with sample preparation errors, beam drift, and instrumental noise corrupting up to one-third of datasets. Automated crystallographic fidelity restoration solutions are needed because corrupted data force engineers to abandon valuable experiments or manually restore datasets at risk of errors. Current image inpainting techniques fail to maintain crystallographic constraints, resulting in restorations that violate the basic rules for crystalline materials. A novel physics-constrained framework is proposed to fill this gap. It integrates adversarial learning with graph neural networks (GNNs) for crystallographically consistent EBSD image inpainting. The proposed GTRG method consists of three elements: i ) a generative adversarial network (GAN) for reconstructing grain boundaries; i i ) a crystallography-guided graph transformer (T) that converts pixel data into orientation-boundary graphs; and i i i ) a regression graph convolutional network (RGCN) that links grain, orientation and boundaries to predict missing crystal orientations. The framework mandates a single orientation per grain and preserves grain boundary structure through structured graph representations. A strategy for creating automated EBSD datasets that incorporates realistic corruption patterns supports effective model training and evaluation. Experimental validation shows better performance than current methods, with a 3.5 % improvement in SSIM (0.950 vs. 0.918) and a 63.0 % reduction in FID (16.55 vs. 44.70) compared to AOT-GAN. The study on aerospace niobium alloys further validates practical utility, showing statistically consistent grain orientation and size distributions (Kolmogorov-Smirnov D = 0.02 , p > 0.98 ). This work introduces two key advancements: 1) the first integration of graph neural networks with adversarial learning for topology-aware image inpainting, and 2) a physics-informed framework bridging computer vision and materials science, enabling effective restoration of corrupted EBSD data for subsequent engineering applications.}, archive = {J_ESWA}, author = {Baiyang Zheng and Jiongran Wen and Yat-Sze Choy and Chengwei Fei}, doi = {10.1016/j.eswa.2025.129667}, journal = {Expert Systems with Applications}, month = {3}, pages = {129667}, shortjournal = {Expert Syst. Appl.}, title = {Physics-constrained EBSD image inpainting via adversarial graph learning: Bridging crystallographic rules and multimodal deep learning}, volume = {298}, year = {2026}, }
• (2026). Three-way large-scale group decision-making under incomplete multi-scale information systems: A perspective of quantum social networks. ESWA, 298, 129666. (www)

  @article{ , abstract = {The gathering and sharing of information lay the groundwork for decision-making, while large-scale group decision-making (LSGDM) strategies address biases, promoting a more comprehensive evaluation of alternatives. Regarding information representation, incomplete multi-scale information systems (MSISs), as an application of granular computing, combine inputs from decision-makers (DMs) and tackle data gaps through multi-level analysis to foster LSGDM. Furthermore, given the interference effect among DMs, quantum social networks (SNs) and three-way decisions (TWD) are vital for effective decision-making. Quantum SNs provide a framework for modeling complex trust relationships among DMs, while TWD offers a structured approach to manage uncertainty. Therefore, this paper seeks to investigate quantum SN-guided three-way LSGDM under incomplete MSISs. First, MSISs are designed to gather information across spatial dimensions. Second, trust propagation paths within SNs are aggregated using quantum theory. Following community clustering through the Leiden algorithm, each community is further divided into core and fringe regions by three-way clustering (TWC), where core alternatives reflect the central members and fringe alternatives represent uncertain members. Third, to achieve intra-group consensus, the weights of DMs in fringe regions and those with low consensus levels are adjusted, while for inter-group consensus, the weight and decision information of community representatives with low consensus levels are modified. Fourth, alternatives are classified using the TWD method, which is grounded in the Dempster-Shafer theory and incorporates the enhanced belief Jensen-Sharma-Mittal ( E B J S M ) divergence. Finally, air quality datasets are used to validate the practicality of this method through sensitivity analysis, simulation analysis, comparative analysis, and statistical analysis.}, archive = {J_ESWA}, author = {Rui Li and Chao Zhang and Hamido Fujita and Wentao Li and Witold Pedrycz and Oscar Castillo}, doi = {10.1016/j.eswa.2025.129666}, journal = {Expert Systems with Applications}, month = {3}, pages = {129666}, shortjournal = {Expert Syst. Appl.}, title = {Three-way large-scale group decision-making under incomplete multi-scale information systems: A perspective of quantum social networks}, volume = {298}, year = {2026}, }
• (2026). GDCR: Geometry-enhanced directional consistency representation for point cloud analysis. ESWA, 298, 129665. (www)

  @article{ , abstract = {Point clouds provide discrete representations of 3D scenes. The relative positions and directions between points collectively describe the objects. Variations in sampling angles, distances, or noises can introduce perturbations, disrupting these spatial and directional relationships. These pose significant challenges for achieving robust feature representations. However, research on the robust representation of point clouds is limited. Although advanced models achieve impressive performance, they exhibit poor robustness to perturbations. To address this issue, we propose Geometry-enhanced Directional Consistency Representation (GDCR), a novel method designed to enhance robustness. In GDCR, we introduce Statistic-based Geometric Reasoning (SGR) to achieve precise spatial geometric estimation for discrete point sets, explicitly enriching spatial geometric information. Furthermore, GDCR vectorizes features embedded with SGR information and applies feature rotation and relative direction refinement in the expanded feature space for robust directional representation. GDCR improves the flexibility and directional expressiveness of point cloud features, significantly improving robustness against perturbations. Extensive experiments demonstrate that GDCR exhibits outstanding robustness while surpassing or matching the performance of state-of-the-art methods.}, archive = {J_ESWA}, author = {Ziming Wang and Boxiang Zhang and Ming Ma and Yue Wang and Taoli Du and Ying Wang and Wenhui Li}, doi = {10.1016/j.eswa.2025.129665}, journal = {Expert Systems with Applications}, month = {3}, pages = {129665}, shortjournal = {Expert Syst. Appl.}, title = {GDCR: Geometry-enhanced directional consistency representation for point cloud analysis}, volume = {298}, year = {2026}, }
• (2026). APMoE-net: Fourier amplitude-phase joint enhancement and MoE compensation for low-light image enhancement. ESWA, 298, 129664. (www)

  @article{ , abstract = {Low-Light Image Enhancement (LLIE) plays a crucial role in computer vision applications. Beyond spatial-based approaches, recent works have explored the Fourier domain. To better preserve structural details in extremely dark scenes, infrared modality has also been introduced as a robust prior for capturing scene geometry. However, existing methods suffer from limited enhancement performance due to the independent modeling of Fourier amplitude and phase, the limitations of cross-modal guidance, and the information loss in sequential feature extraction. To address these challenges, we propose APMoE-Net, a dual-stage Fourier network framework with amplitude-phase joint enhancement and spatial Mixture of Experts (MoE) compensation. Stage one performs coarse enhancement by leveraging infrared images to jointly optimize Fourier amplitude and phase, enabling mutual guidance learning between them. Subsequently, a Modality Refinement Module leverages edge information to refine infrared inputs, producing a refined modality map as a more accurate cross-modal prior for subsequent processing. The second stage employs a dual-branch design for texture refinement. Our key innovation lies in the MoE Compensation Module integrated within the Multi-scale Convolution Branch. This module employs a dynamic routing network to selectively activate specialized experts, enabling the recovery of fine-grained textures that are lost during sequential processing. Meanwhile, the Fourier Branch integrates the refined modality map to improve overall detail and contrast. Comprehensive experiments demonstrate that APMoE-Net surpasses state-of-the-art (SOTA) methods in both qualitative and quantitative evaluations. Notably, APMoE-Net achieves outstanding performance with a lightweight design, offering an efficient LLIE solution.}, archive = {J_ESWA}, author = {Mengen Cai and Tongshun Zhang and Pingping Liu and Qiuzhan Zhou}, doi = {10.1016/j.eswa.2025.129664}, journal = {Expert Systems with Applications}, month = {3}, pages = {129664}, shortjournal = {Expert Syst. Appl.}, title = {APMoE-net: Fourier amplitude-phase joint enhancement and MoE compensation for low-light image enhancement}, volume = {298}, year = {2026}, }
• (2026). Enhanced edge detection of harmful algal blooms using diffusion probability models and sobel-convolutional attention mechanisms. ESWA, 298, 129663. (www)

  @article{ , abstract = {Accurate detection and identification of harmful algal bloom (HAB) images are crucial for developing effective early warning systems for HABs. However, existing edge detection models, primarily designed for natural scenes, struggle with HAB-specific challenges such as blurred cell contours and interference from impurity bubbles. To address these issues, we propose a novel edge detection approach tailored for marine HABs, integrating a diffusion probability model with Sobel convolutional inter-layer attention mechanisms. Firstly, we develop an image enhancement algorithm specifically for HABs images, significantly improving real-time dynamic sampling data by enhancing contrast, edges, and texture features. Next, we introduce the SIAnet network, which utilizes inter-layer attention and convolutional operations to generate comprehensive global information. This network enhances feature correlation by aggregating shared features across multiple layers and modeling both long-range and short-range dependencies, effectively suppressing noise and background interference. This facilitates precise extraction of algae boundaries and morphological characteristics. Additionally, an improved Sobel operator is employed to generate supplementary edge features, accelerating the training process. Experimental results demonstrate that the proposed method achieves robust performance on the HABs dataset, with an Optimal Dataset Scale (ODS) of 0.645, an Optimal Image Scale (OIS) of 0.702, and an Average Precision (AP) of 0.813. Compared to existing methodologies, our approach demonstrates strong generalization on BSDS and BIPED datasets, significantly enhancing performance and mitigating typical CNN issues of edge thickening and fragmentation. It offers essential technical support for efficient HAB early warning system development.}, archive = {J_ESWA}, author = {Gengkun Wu and Yining Fan and Xin Tian and Chao Cui and Jiazheng Han}, doi = {10.1016/j.eswa.2025.129663}, journal = {Expert Systems with Applications}, month = {3}, pages = {129663}, shortjournal = {Expert Syst. Appl.}, title = {Enhanced edge detection of harmful algal blooms using diffusion probability models and sobel-convolutional attention mechanisms}, volume = {298}, year = {2026}, }
• (2026). Predictive modeling of biogeographical ancestry using a novel SNP panel and supervised learning approaches. ESWA, 298, 129662. (www)

  @article{ , abstract = {Inferring an individual’s BioGeographical Ancestry (BGA) through DNA analysis is a valuable tool in various fields such as forensic science, especially when traditional methods fail to identify suspects or victims. Advances in Next-Generation Sequencing (NGS) have revolutionized genomic data acquisition, enabling the development of comprehensive Single Nucleotide Polymorphism (SNP) panels for ancestry inference. This study assessed the effectiveness of a novel panel containing 3234 SNPs at both inter-continental and a more detailed BGA level, using various supervised Machine Learning (ML) models, including Categorical Naive Bayes, Penalized Multinomial Logistic Regression, Linear Support Vector Machines, Random Forest, and tree-based Gradient Boosting. A nested cross-validation approach was employed for model tuning and evaluation, with balanced accuracy as the main performance metric to address class imbalance. At the inter-continental level, all ML models demonstrated high balanced accuracy, confirming their reliability for BGA inference. However, performance declined at the more detailed continental level, likely due to a combination of factors including increased class imbalance, reduced sample sizes for certain populations, and the inherent complexity of distinguishing genetically and geographically proximate groups. Nonetheless, promising results were observed for South Asians, Northeast Asians, Europeans, and West Africans classes. In contrast, performance was notably lower for underrepresented classes such as Inner Asians. Misclassification patterns at both levels appeared to reflect known geographical and historical relationships, although further analysis revealed that these were often concentrated in underrepresented or genetically complex groups. These findings highlighted the potential of this SNP panel and ML approaches as valuable tools for forensic investigations.}, archive = {J_ESWA}, author = {Cosimo Grazzini and Giorgia Spera and Stefania Morelli and Daniele Castellana and Giulia Cosenza and Michela Baccini and Giulia Cereda and Elena Pilli}, doi = {10.1016/j.eswa.2025.129662}, journal = {Expert Systems with Applications}, month = {3}, pages = {129662}, shortjournal = {Expert Syst. Appl.}, title = {Predictive modeling of biogeographical ancestry using a novel SNP panel and supervised learning approaches}, volume = {298}, year = {2026}, }
• (2026). RL-CoSeg: A reinforcement learning-based collaborative localization and segmentation framework for medical image. ESWA, 298, 129661. (www)

  @article{ , abstract = {Segmenting small regions of interest (ROIs) from abdominal CT images presents significant challenges, particularly due to class imbalance and variations in the sizes of foreground objects. A commonly adopted solution is the two-stage segmentation. However, this approach has two key limitations: i) Difficulty in balancing localization accuracy and target preservation. To reduce information loss in the first stage, existing methods typically enlarge the predicted bounding boxes, which improves coverage but compromises localization precision. ii) Independent optimization of the two stages, which lacks a collaborative mechanism. This fragmented pipeline limits the flow of information between stages, thereby constraining performance improvements. To address these limitations, we propose a reinforcement learning-based collaborative localization and segmentation (RL-CoSeg) framework, which comprises three sub-networks: localization network (LN), segmentation network, and localization-segmentation collaboration network (LSCN). The LN integrates prior knowledge and incorporates a dynamic reward mechanism to enhance the accuracy and efficiency of target detection through reinforcement learning (RL) strategies. The LSCN further introduces segmentation predictions as a reward signal, which, together with the localization reward, jointly drives policy learning. In addition, a heuristic exploration strategy is employed to avoid local optima and improve training stability. This design strengthens the information interaction and collaborative performance between the two tasks. Experimental results demonstrate that the proposed method achieves superior collaborative performance in small-target medical image segmentation.}, archive = {J_ESWA}, author = {Feilong Xu and Feiyang Yang and Xiaoli Zhang and Zhaojun Liu}, doi = {10.1016/j.eswa.2025.129661}, journal = {Expert Systems with Applications}, month = {3}, pages = {129661}, shortjournal = {Expert Syst. Appl.}, title = {RL-CoSeg: A reinforcement learning-based collaborative localization and segmentation framework for medical image}, volume = {298}, year = {2026}, }
• (2026). SH-ETRs:Learning soft-hard rules with entity type constraints for document-level relation extraction. ESWA, 298, 129660. (www)

  @article{ , abstract = {Document-level relation extraction (DocRE) aims to extract relations between entity pairs across the entirety of a document. Current methods have begun to adopt logical rules to enhance the performance of DocRE models. However, the pipeline’s rule learning framework will suffer from the issue of error propagation, and the end-to-end method may lead to mistakes in rule reasoning. Additionly, they ignore entity type information when learning the rules. To address these issues, we propose a novel framework named Soft-Hard Rules with Entity Type Constraints (SH-ETRs) for improving the rules’ expressiveness and quality. Specifically, we first propose a Hard Entity Type Rules Module (H-ETRs) to learn entity type information and provide hard rule constraints. Then, we propose a Soft Entity Type Rule Reasoning Module (S-ETRs), which parameterizes the rule inference process and reduces error propagation during the process. Furthermore, by applying a rule consistency loss function to S-ETRs, we achieve the learning of soft rules under hard rule constraints, thereby aiming to prevent the learning of inaccurate rules during the training process. The experimental results demonstrate that our method outperforms existing rule learning frameworks, achieving state-of-the-art performance with an F1 score of 74.39 and an IgnF1 score of 67.43 across three public datasets and two baseline models.}, archive = {J_ESWA}, author = {Haisong Chen and Nisuo Du and Qing He and Yuji Wang}, doi = {10.1016/j.eswa.2025.129660}, journal = {Expert Systems with Applications}, month = {3}, pages = {129660}, shortjournal = {Expert Syst. Appl.}, title = {SH-ETRs:Learning soft-hard rules with entity type constraints for document-level relation extraction}, volume = {298}, year = {2026}, }
• (2026). Interpretable knowledge tracing with dual-level knowledge states. ESWA, 298, 129658. (www)

  @article{ , abstract = {Knowledge Tracing (KT) is a critical technology for achieving personalized learning. It estimates learners’ knowledge states and predicts future performance using historical interaction data. Despite recent advances, two significant challenges remain. First, the accuracy of knowledge state modeling is limited by the insufficient fusion of multi-scale information across temporal and spatial dimensions. Second, a trade-off persists between improving predictive performance and enhancing interpretability. This paper proposes an interpretable knowledge tracing method based on dual-level knowledge states (DIKT) to address these challenges. From the temporal perspective, DIKT incorporates a forgetting-aware RoLinear Transformer and a semantic similarity-based review mechanism to model learners’ problem-level knowledge states. From the spatial perspective, it leverages a Knowledge Concept (KC) relational graph to propagate influence among related KCs and dynamically update learners’ concept-level knowledge states through three sequential learning phases: forgetting, aggregation, and updating. Student performance is predicted using a two-parameter Item Response Theory (IRT) model, which incorporates guess and slip parameters to account for response anomalies. We conduct extensive comparisons between DIKT and 20 state-of-the-art KT models on five widely used public datasets. Experimental results demonstrate that DIKT achieves superior performance while preserving interpretability, highlighting its practical potential for real-world educational applications. The code is available at https://github.com/ting214/DIKT .}, archive = {J_ESWA}, author = {Yanting Li and Tao Zhou and Tianyu Cai and Shenggen Ju}, doi = {10.1016/j.eswa.2025.129658}, journal = {Expert Systems with Applications}, month = {3}, pages = {129658}, shortjournal = {Expert Syst. Appl.}, title = {Interpretable knowledge tracing with dual-level knowledge states}, volume = {298}, year = {2026}, }
• (2026). RASpan: Improving toponym recognition through span representation model with retrieval augmentation. ESWA, 298, 129657. (www)

  @article{ , abstract = {Toponym recognition aims to identify place names from natural language texts, which is vital for various applications including geographic information retrieval, emergency response, and natural disaster analysis. Currently, mainstream studies mainly adopt deep learning models for toponym recognition. However, these approaches encounter significant limitations due to the inherent ambiguity, variation, and abbreviation of toponyms. To address these issues, we propose a novel Span Representation Model with R etrieval A ugmentation ( RASpan ) that leverages more accurate span representation and effective external geo-entity information to enhance the semantic representation of place names for improving the performance of toponym recognition. On the one hand, RASpan retrieves diverse geo-entities and concatenates geo-entity knowledge with an input sequence to construct a new prompt sequence. On the other hand, RASpan utilizes the prompt encoder based on the language model to encode this prompt sequence and employs a dedicated span representation module to obtain more accurate span representations. In addition, a new geo-entity prediction task is designed to learn the entire representation of each geo-entity while minimizing noise interference. Experiments on three publicly available datasets demonstrate that our model achieves new state-of-the-art results, highlighting the effectiveness of RASpan in toponym recognition by introducing prior geo-entity knowledge.}, archive = {J_ESWA}, author = {Hui Wu and Anran Yang and Zhinong Zhong and Ye Wu and Fei Yang and Luo Chen and Ning Jing}, doi = {10.1016/j.eswa.2025.129657}, journal = {Expert Systems with Applications}, month = {3}, pages = {129657}, shortjournal = {Expert Syst. Appl.}, title = {RASpan: Improving toponym recognition through span representation model with retrieval augmentation}, volume = {298}, year = {2026}, }
• (2026). An interpretable automated optimized machine learning for predicting concrete compressive strength. ESWA, 298, 129656. (www)

  @article{ , abstract = {This study presents a novel, interpretable, and automated machine learning (AutoML) framework for accurately predicting the compressive strength of environmentally sustainable concrete mixtures that incorporate supplementary cementitious materials (SCMs) by addressing the growing need for transparent and data driven tools in structural material design, particularly for concrete mixes enriched with various SCMs. A robust unified dataset of 1,317 samples was curated by integrating peer-reviewed experimental studies for this study. The proposed methodology incorporates feature contribution ranking through mutual information, model screening with AutoML to identify the most effective regression models, Bayesian optimization for fine-tuning model parameters, and interpretability techniques including SHAP and counterfactual analysis. The best performance metrics, in training include R 2 of 0.999, a mean absolute error 0.114, root mean squared error 0.7094, and mean absolute percentage error of 0.51 %, in testing phase R 2 of 0.944, a mean absolute error 3.479, root mean squared error 4.8173, and mean absolute percentage error of 9.86 %. The weakest performance, with a training R 2 of 0.982 and a mean absolute error of 1.911 MPa, root mean squared error 2.7990 and mean absolute percentage error 5.43 %, in testing phase R 2 of 0.786 and a mean absolute error of 5.754 MPa, root mean squared error 9.4336 and mean absolute percentage error 16.16 %. The interpretability analysis values provided insights into the most important features, such as curing time and cement are crucial in predicting the strength. Counterfactual analysis further validated the model by illustrating the significant impact of cement, age and water on concrete strength.}, archive = {J_ESWA}, author = {Aparna Kamarthi and Baskar Kaliyamoorthy}, doi = {10.1016/j.eswa.2025.129656}, journal = {Expert Systems with Applications}, month = {3}, pages = {129656}, shortjournal = {Expert Syst. Appl.}, title = {An interpretable automated optimized machine learning for predicting concrete compressive strength}, volume = {298}, year = {2026}, }
• (2026). The power of text similarity in identifying AI-LLM paraphrased documents: The case of BBC news articles and ChatGPT. ESWA, 298, 129655. (www)

  @article{ , abstract = {Generative AI paraphrased text can be used for copyright infringement and the AI paraphrased content can deprive substantial revenue from original content creators. Despite this recent surge of malicious use of generative AI, there are few academic publications that research this threat. In this article, we demonstrate the ability of pattern-based similarity detection for AI paraphrased news recognition. We propose an algorithmic scheme, which is not limited to detect whether an article is an AI paraphrase, but, more importantly, to identify that the source of infringement is the ChatGPT. The proposed method is tested with a benchmark dataset specifically created for this task that incorporates real articles from BBC, incorporating a total of 2,224 articles across five different news categories, as well as 2,224 paraphrased articles created with ChatGPT. Results show that our pattern similarity-based method, that makes no use of deep learning, can detect ChatGPT assisted paraphrased articles at percentages 96.23% for accuracy, 96.25 for precision, 96.21% for sensitivity, 96.25% for specificity and 96.23% for F1 statistic.}, archive = {J_ESWA}, author = {Konstantinos F. Xylogiannopoulos and Petros Xanthopoulos and Panagiotis Karampelas and Georgios A. Bakamitsos}, doi = {10.1016/j.eswa.2025.129655}, journal = {Expert Systems with Applications}, month = {3}, pages = {129655}, shortjournal = {Expert Syst. Appl.}, title = {The power of text similarity in identifying AI-LLM paraphrased documents: The case of BBC news articles and ChatGPT}, volume = {298}, year = {2026}, }
• (2026). Dynamic collaborative evolutionary network: A novel spatio-temporal feature extraction framework for EEG emotion recognition. ESWA, 298, 129654. (www)

  @article{ , abstract = {In recent years, significant progress has been made in emotion recognition research based on electroencephalogram (EEG) signals. However, existing methods face two key limitations: on one hand, the reliance on fixed physical connections or static topological relationships makes it difficult to effectively represent the dynamic non-Euclidean spatial characteristics between EEG electrodes; on the other hand, spatiotemporal feature extraction is often conducted independently. This lack of a collaborative mechanism for spatiotemporal features results in insufficient fine-grained emotional representation capability. To address these issues, a dynamic collaborative evolutionary network (DCENet) is proposed based on graph-aware enhancement and global convolutional Transformer for EEG emotion recognition. DCENet constructs the causal relationship between electrodes by constructing the graph-aware enhancement (GAE) module, obtains spatial features with the causal relationship, and enhances key features. At the same time, DCENet constructs the global convolutional Transformer (GCT) module, which utilizes the global modeling advantage of the Transformer and the local perception ability of the convolutional operation to capture the temporal features with different scales. In addition, DCENet adaptively fuses temporal and spatial features through the local differential fusion (LDF) module to achieve cross-domain feature alignment and feature alignment of emotion categories to collaboratively evolve emotion representation features with more fine-grained information. This paper conducts experiments on the SEED, SEED-IV, and MPED datasets to validate the effectiveness of DCENet. The experimental results show that the model achieves cross-subject average accuracies of 87.55 %, 73.04 %, and 27.72 % on SEED, SEED-IV, and MPED, respectively, outperforming the state-of-the-art methods. The source code is publicly available at: https://github.com/cvmdsp/DCENet .}, archive = {J_ESWA}, author = {Shuaiqi Liu and Zhihui Gu and Yuan Zhang and Yanling An and Shuhuan Zhao and Bing Li and Yudong Zhang}, doi = {10.1016/j.eswa.2025.129654}, journal = {Expert Systems with Applications}, month = {3}, pages = {129654}, shortjournal = {Expert Syst. Appl.}, title = {Dynamic collaborative evolutionary network: A novel spatio-temporal feature extraction framework for EEG emotion recognition}, volume = {298}, year = {2026}, }
• (2026). Trust in recommender systems: A survey. ESWA, 298, 129653. (www)

  @article{ , abstract = {Trust-based recommender systems incorporate interpersonal trust relationships into the recommendation process, operating on the principle that users are more likely to accept suggestions from people they trust. Empirical studies have shown that trust-aware approaches often deliver more accurate recommendations than their trust-unaware counterparts. In this comprehensive, up-to-date survey, we analyze a variety of trust-based recommendation methods, categorize different trust inference techniques, and examine how trust is integrated into recommendation algorithms. We then organize the investigated approaches according to a clear taxonomy, explore their underlying concepts, and highlight the key challenges and open issues that remain in the field.}, archive = {J_ESWA}, author = {Imane Akdim and Loubna Mekouar and Youssef Iraqi}, doi = {10.1016/j.eswa.2025.129653}, journal = {Expert Systems with Applications}, month = {3}, pages = {129653}, shortjournal = {Expert Syst. Appl.}, title = {Trust in recommender systems: A survey}, volume = {298}, year = {2026}, }
• (2026). Composite human activity recognition utilizing knowledge distillation and sensor fusion focusing on resource constrained microcontrollers. ESWA, 298, 129652. (www)

  @article{ , abstract = {This study presents a cost-effective, low-computation system for composite Human Activity Recognition (HAR) that leverages knowledge-distilled neural networks on a Microcontroller Unit (MCU) to minimize reliance on cloud processing. A key contribution of this work is the investigation of plantar pressure sensor data within a knowledge distillation framework, addressing a notable gap in the existing literature. The proposed solution centers around the ESP32-S3 DevKit C1, equipped with a dual-core 240 MHz Tensilica chip, 320 KiB of usable Static Random Access Memory (SRAM), and built-in Wi-Fi and Bluetooth. Significantly, both the teacher and the student models surpass existing state-of-the-art methods, achieving F1-scores of 99.33 %, 98.36 %, and 97.68 % respectively, in classifying a comprehensive set of 21 activities (15 composite and 6 simple). The distilled student models demonstrate remarkable efficiency, with execution times of 1.83 and 0.64 s, memory footprints of only 62 KB and 82 KB, and flash memory usage of approximately 209 KB and 127 KB, while maintaining low power consumption of 210 mW and 215 mW, respectively. Furthermore, we have developed an end-to-end prototype that integrates the ESP32-S3 with a WitMotion Inertial Measurement Unit (IMU) sensor. This system autonomously manages data acquisition, feature extraction, and inference in under 7 s with a total power consumption of approximately 295 mW.}, archive = {J_ESWA}, author = {Athar Noor Mohammad Rafee and John Clear and Jannatun Noor}, doi = {10.1016/j.eswa.2025.129652}, journal = {Expert Systems with Applications}, month = {3}, pages = {129652}, shortjournal = {Expert Syst. Appl.}, title = {Composite human activity recognition utilizing knowledge distillation and sensor fusion focusing on resource constrained microcontrollers}, volume = {298}, year = {2026}, }
• (2026). Med-D3CG: Wavelet-based diffusion in the difference domain for cross-modality medical image generation. ESWA, 298, 129651. (www)

  @article{ , abstract = {The synthesis of cross-modal medical images plays a vital role in bridging diagnostic gaps between imaging modalities such as CT, MRI, and PET. This integration enables a more comprehensive evaluation of a patient’s condition, improving diagnostic accuracy and aiding clinical decision-making. However, the performance of conditional denoising diffusion probabilistic models is often hindered by pronounced structural and intensity discrepancies between modalities, as well as the inherently slow nature of the diffusion process. To address these challenges, this paper proposes Wavelet-Based Diffusion in the Difference Domain for Cross-Modality Medical Image Generation (Med-D3CG), a novel framework that transforms the synthesis process by emphasizing the difference domain. Instead of directly generating target images like conventional methods, Med-D3CG models the residual information between conditioned and target images. This strategy allows the framework to accurately capture essential structural and intensity variations between modalities, leading to more precise and realistic image synthesis. Additionally, Med-D3CG integrates the Discrete Wavelet Transform (DWT) to improve efficiency, accelerating the diffusion process while maintaining high image fidelity. On SynthRAD2023 and HMIFD datasets, state-of-the-art performance is achieved on pelvis and HMIFD using Med-D3CG , with the best Learned Perceptual Image Patch Similarity (LPIPS) and competitive FID observed on brain. Code and pretrained models are provided at https://github.com/ZgzTTTer/Med-D3CG .}, archive = {J_ESWA}, author = {Guangzhen Zhu and Midi Wan and Wenming Cao and Zhiwen Yu and Jin Hu and Bing Li and Xiaotao Fan}, doi = {10.1016/j.eswa.2025.129651}, journal = {Expert Systems with Applications}, month = {3}, pages = {129651}, shortjournal = {Expert Syst. Appl.}, title = {Med-D3CG: Wavelet-based diffusion in the difference domain for cross-modality medical image generation}, volume = {298}, year = {2026}, }
• (2026). Two-stage feature selection utilizing three-way adaptive neighborhood characteristic measure and optimal combination search. ESWA, 298, 129650. (www)

  @article{ , abstract = {The neighborhood rough set model(NRSM) has shown its powerful capacity in feature selection. However, a challenge still exists in describing the diversity between the attributes deeply while avoiding the impact caused by the neighborhood parameters. To address this problem, in this paper, we propose a two-stage feature selection by utilizing a three-way adaptive characteristic measure and an optimal combination search. First, we define a fitness function for applying the Stochastic Fractal Search(SFS) to design an novel adaptive neighborhood rough set model(ANRSM). To better utilize the construction characteristic of the adaptive model and reduce the computational cost, the lower and upper approximations of the SFS-based ANRSM are redefined through the fitness function. Second, based on the two approximations, we analyze the fitness function thresholds that can partition the universe into three regions and design the three-way adaptive neighborhood characteristic regions, which provide a more direct classification of samples without the inclusion and union operations. Third, we design different measures for the samples in diverse regions based on their corresponding characteristic. Afterward, a three-way adaptive characteristic measure is designed by integrating the three measures to evaluate the uncertainty of attributes. Then, we apply the measure to design a feature selection approach with greedy search. Considering that the greedy strategy may output redundant attributes, we introduce an optimal combination search approach through a novel wrapper technology to explore the potential optimal feature combinations. Compared with nine algorithms on fourteen public datasets, the experimental results show the effectiveness of our algorithm.}, archive = {J_ESWA}, author = {Bowen Lin and Duoqian Miao and Caihui Liu and Hongyun Zhang and Ruizhi Wang and Witold Pedrycz}, doi = {10.1016/j.eswa.2025.129650}, journal = {Expert Systems with Applications}, month = {3}, pages = {129650}, shortjournal = {Expert Syst. Appl.}, title = {Two-stage feature selection utilizing three-way adaptive neighborhood characteristic measure and optimal combination search}, volume = {298}, year = {2026}, }
• (2026). New semi-supervised fuzzy C-means clustering with asymmetric deviation constraints and fast algorithm. ESWA, 298, 129648. (www)

  @article{ , abstract = {Semi-supervised clustering leverages prior information to improve algorithm performance and is widely valued by researchers. This paper analyzes the traditional semi-supervised fuzzy C-means (SFCM) objective function, noting that as a labeled sample’s membership degree aligns with its prior information, the impact of this information on the deviation constraint weakens. This reduces its supervisory effect on optimizing the membership partition matrix, especially with a large regularization factor. To overcome this, we propose a novel semi-supervised fuzzy C-means method based on an asymmetric deviation constraint and develop a two-level alternating iterative optimization algorithm, supported by theoretical convergence analysis using Zangwill’s theorem and the bordered Hessian matrix. To address the slow convergence and high computational cost typical of semi-supervised fuzzy clustering, we further enhance the algorithm with affinity filtering and a membership scaling scheme for improved efficiency. Experimental results demonstrate that our methods significantly outperform existing state-of-the-art techniques, advancing semi-supervised fuzzy C-means clustering.}, archive = {J_ESWA}, author = {Chengmao Wu and Jun Hou}, doi = {10.1016/j.eswa.2025.129648}, journal = {Expert Systems with Applications}, month = {3}, pages = {129648}, shortjournal = {Expert Syst. Appl.}, title = {New semi-supervised fuzzy C-means clustering with asymmetric deviation constraints and fast algorithm}, volume = {298}, year = {2026}, }
• (2026). BayesAHDD: A new bayesian rule-based adaptive hypersphere data description for few-shot one-class classification. ESWA, 298, 129647. (www)

  @article{ , abstract = {Few-shot one-class classification (FS-OCC) is a challenging classification problem that involves learning from a very limited number of training samples, all from a single class. Recently, several data description methods have been proposed to address the FS-OCC problem. Unlike conventional one-class classification problems, the few-shot setting requires the model to generalize to novel tasks with previously unseen positive classes. Most existing methods learn decision boundaries in the feature space without explicitly modeling the underlying data distributions, which limits the generalization ability of the learned representations. To address this issue, we propose Bayesian Rule-based Adaptive Hypersphere Data Description (BayesAHDD), a probabilistic framework that represents data with multivariate Gaussian distributions and performs classification according to the Bayes decision rule. Based on the assumption that negative samples are more dispersed in the feature space, BayesAHDD models the negative class by scaling the positive class variance vector element-wise using a learnable vector. To address the challenges of exploding gradients and numerical overflow, we impose a lower bound on the positive class variance vector and introduce a trainable parameter that integrates the class prior probability ratio with the normalization constants of the Gaussian class-conditional densities. Experimental results on both benchmark and domain-specific datasets show that BayesAHDD consistently outperforms existing baselines and state-of-the-art FS-OCC methods. Moreover, quantitative analysis demonstrates that the learned feature representations exhibit superior discriminative ability compared to those produced by previous approaches.}, archive = {J_ESWA}, author = {Yuchen Ren and Xiabi Liu and Yan Pei and Yunlong Li and Yongxia Wei}, doi = {10.1016/j.eswa.2025.129647}, journal = {Expert Systems with Applications}, month = {3}, pages = {129647}, shortjournal = {Expert Syst. Appl.}, title = {BayesAHDD: A new bayesian rule-based adaptive hypersphere data description for few-shot one-class classification}, volume = {298}, year = {2026}, }
• (2026). A network selection algorithm for space-air-ground integrated network based on location prediction model and multi-attribute decision making. ESWA, 298, 129646. (www)

  @article{ , abstract = {As an indispensable component of 5G and even the future 6G networks, the Space-Air-Ground Integrated Network (SAGIN) is envisioned to provide ubiquitous network connectivity and services by integrating satellite, aerial, and terrestrial networks. However, due to the frequent network selection of in-vehicle terminals, the user’s Quality of Service (QoS) can significantly deteriorate. To address this issue, a network selection algorithm based on terminal location prediction has been proposed. Firstly, we enhanced the Particle Swarm Optimization (PSO) algorithm to optimize the hyper-parameters of the Long Short-Term Memory (LSTM) network, thereby improving the accuracy of terminal location prediction. After constructing the network sets of the current terminal position and the predicted position, respectively, we designed a network selection judgment mechanism with a dynamically adjustable switching threshold based on Fuzzy Logic and K-Means theory. Finally, through the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) algorithm, we have achieved robust network selection in fast-moving scenarios. The simulation results show that the proposed algorithm can adaptively adjust the switching threshold and provide precise positions. Compared to existing algorithms, it can significantly reduce the number of candidate networks and the number of selections, thereby reducing the computational load and increasing the throughput of users.}, archive = {J_ESWA}, author = {Jianli Xie and Weicheng Pan and Lei Wu and Zishan Wu}, doi = {10.1016/j.eswa.2025.129646}, journal = {Expert Systems with Applications}, month = {3}, pages = {129646}, shortjournal = {Expert Syst. Appl.}, title = {A network selection algorithm for space-air-ground integrated network based on location prediction model and multi-attribute decision making}, volume = {298}, year = {2026}, }
• (2026). Anomaly detection based on graph neural networks incorporating with domain knowledge for industrial cyber-physical systems. ESWA, 298, 129645. (www)

  @article{ , abstract = {Effective anomaly detection heavily relies on accurately modeling the normal behavior of the industrial cyber-physical systems (ICPSs). Current popular data-driven black-box methods’ performance depends on the quantity and quality of data from the ICPS, rather than integrating with the intrinsic characteristics of the system, such as mass conservation and structural dependencies of industrial processes. The data used to train these models often fails to cover all operating conditions of ICPSs, resulting in a distributional mismatch between training and testing data. This limitation significantly reduces the generalization capability of trained data-driven models for detecting anomalies under unseen and unknown operational conditions. Meanwhile, constructing well-generalized behavior models using white-box mechanism models (MM)–which represent the internal behavior of controllers and physical processes using first-principles equations or physical laws, such as conservation principles and kinetic dynamics–is not always feasible in real-world industrial applications. To address this issue, this paper presents a novel anomaly detection approach employing a domain knowledge-embedded hybrid graph neural network (HGNN) that integrates control, physical, and structural characteristics of ICPSs into a data-driven modeling framework. It establishes a domain knowledge-based MM to predict the behavior of controllers and physical processes, and utilizes a HGNN model with embedded state topology and spatial distribution knowledge to compensate for the predictive error of MM, forming a hybrid model called mechanism-embedded HGNN (MEHGNN). This model can detect anomalies in unknown operational conditions by analyzing the predictive error of MEHGNN. In the experiments, MEHGNN raises the detection accuracy of GNNs from 36.6 %–82.4 % to 79.4 %–96.3 % under distribution shift scenarios, achieving the best detection performance among all compared methods.}, archive = {J_ESWA}, author = {Xin Du and Chunjie Zhou and Yu-Chu Tian and Bo Xu}, doi = {10.1016/j.eswa.2025.129645}, journal = {Expert Systems with Applications}, month = {3}, pages = {129645}, shortjournal = {Expert Syst. Appl.}, title = {Anomaly detection based on graph neural networks incorporating with domain knowledge for industrial cyber-physical systems}, volume = {298}, year = {2026}, }
• (2026). Predictive modeling and multi-objective optimization of screw whirling milling based on ISSA-BP embedded NSGA-III algorithm. ESWA, 298, 129644. (www)

  @article{ , abstract = {In screw whirling milling, the relationship between machining quality and processing parameters exhibits highly nonlinear characteristics. The traditional multiple regression models may not be able to capture this complex relationship accurately. Therefore, it is necessary to consider more flexible and applicable algorithms to establish their connections and optimize processing parameters. It can improve the accuracy and reliability of products, and provide more scientific method guidance for screw whirling milling processing. The originality of this article lies in proposing an adaptive dynamic optimization hybrid model. This model combines improved sparrow search algorithm optimized backpropagation (ISSA-BP) and non-dominated sorting genetic algorithm (NSGA-III). It can effectively adapt to dynamic data and find the optimal balance point among multiple objectives to better predict and optimize responses (cutting force, vibration, roughness, and residual compressive stress) in screw whirling milling. Firstly, a suitable network structure is identified by comparing the effects of five improvement strategies, population size, and the ratio of producers to scouters on the sparrow search algorithm. Then, an ISSA-BP prediction model is developed for four responses based on this structure. On this basis, the superiority of the established ISSA-BP model is verified by comparing prediction performance of five algorithms, and the relative prediction errors are all within 2%. The R 2 values of the models are all above 0.99, and they also perform well in indicators such as MAE (Mean Absolute Error), MSE (Mean Squared Error), MAPE (Mean Absolute Percentage Error), and RMSE (Root Mean Squared Error). Then, ISSA-BP model is encapsulated and embedded into the optimization algorithm as the fitness function of NSGA-III. Finally, with the processing parameters of whirling milling as constraints, the NSGA-III algorithm is used to solve the proposed model and obtain the Pareto optimal solution set. Choosing appropriate processing parameters according to different needs in actual machining can help improve the quality and efficiency of screw machining.}, archive = {J_ESWA}, author = {Chao Liu and Hao Ding and Juanjuan Zheng and Yan He and Shaofu Huang and Junbo Tuo and Zuqing Luo and Gang Shen}, doi = {10.1016/j.eswa.2025.129644}, journal = {Expert Systems with Applications}, month = {3}, pages = {129644}, shortjournal = {Expert Syst. Appl.}, title = {Predictive modeling and multi-objective optimization of screw whirling milling based on ISSA-BP embedded NSGA-III algorithm}, volume = {298}, year = {2026}, }
• (2026). A survey of large language models for data challenges in graphs. ESWA, 298, 129643. (www)

  @article{ , abstract = {Graphs are a widely used paradigm for representing non-Euclidean data, with applications ranging from social network analysis to biomolecular prediction. While graph learning has achieved remarkable progress, real-world graph data presents a number of challenges that significantly hinder the learning process. In this survey, we focus on four fundamental data-centric challenges: (1) Incompleteness , real-world graphs have missing nodes, edges, or attributes; (2) Imbalance , the distribution of the labels of nodes or edges and their structures for real-world graphs are highly skewed; (3) Cross-domain Heterogeneity , graphs from different domains exhibit incompatible feature spaces or structural patterns; and (4) Dynamic Instability , graphs evolve over time in unpredictable ways. Recently, Large Language Models (LLMs) offer the potential to tackle these challenges by leveraging rich semantic reasoning and external knowledge. This survey focuses on how LLMs can address four fundamental data-centric challenges in graph-structured data, thereby improving the effectiveness of graph learning. For each challenge, we review both traditional solutions and modern LLM-driven approaches, highlighting how LLMs contribute unique advantages. Finally, we discuss open research questions and promising future directions in this emerging interdisciplinary field. To support further exploration, we have curated a repository of recent advances on graph learning challenges: https://github.com/limengran98/Awesome-Literature-Graph-Learning-Challenges .}, archive = {J_ESWA}, author = {Mengran Li and Pengyu Zhang and Wenbin Xing and Yijia Zheng and Klim Zaporojets and Junzhou Chen and Ronghui Zhang and Yong Zhang and Siyuan Gong and Jia Hu and Xiaolei Ma and Zhiyuan Liu and Paul Groth and Marcel Worring}, doi = {10.1016/j.eswa.2025.129643}, journal = {Expert Systems with Applications}, month = {3}, pages = {129643}, shortjournal = {Expert Syst. Appl.}, title = {A survey of large language models for data challenges in graphs}, volume = {298}, year = {2026}, }
• (2026). Region-aware prediction strategy based on shared points and multiple scales for dynamic multi-objective optimization. ESWA, 298, 129642. (www)

  @article{ , abstract = {In dynamic multi-objective optimization problems, effectively predicting and tracking the Pareto optimal front (POF) under environmental changes has been one of the core challenges. In this paper, we propose a region-aware prediction strategy based on shared points and multiple scales (RADMOEA) that combines global and local characteristics, aiming to enhance the algorithm’s ability to sense and adapt to POF. Firstly, the center-point movement strategy is used to move the non-dominated solution set from the previous moment to obtain the non-dominated solution set after the movement. The actual non-dominated solution set at the current moment and the non-dominated solution set after the movement share points in the objective space, and these shared points divide the non-dominated solution set at the current moment into several subregions. Within each region, all individuals are appropriately rescaled, and a local coordinate system is established. Then, within the local coordinate system, each individual is associated with the nearest post-movement non-dominated individual. Finally, new populations adapted to environmental changes are generated by combining centroid movement directions, Gaussian perturbations, and multi-scale individual association relationships. The proposed strategy is compared with six advanced algorithms, and the experimental results demonstrate that RADMOEA is effective in tracking the POF under dynamic environments.}, archive = {J_ESWA}, author = {Yaru Hu and Sitong Wang and Junwei Ou and Zhenlin Mei and Juan Zou and Shengxiang Yang}, doi = {10.1016/j.eswa.2025.129642}, journal = {Expert Systems with Applications}, month = {3}, pages = {129642}, shortjournal = {Expert Syst. Appl.}, title = {Region-aware prediction strategy based on shared points and multiple scales for dynamic multi-objective optimization}, volume = {298}, year = {2026}, }
• (2026). A transformer network for multi-dimensional nonuniform aperture synthesis radiometer image inversion. ESWA, 298, 129641. (www)

  @article{ , abstract = {Passive microwave remote sensing plays a vital role in Earth observation, with applications in soil moisture, ocean salinity, and atmospheric monitoring. However, improving spatial resolution at low frequencies remains challenging. Recently, combining multiple small antenna arrays into a larger one has emerged as a technological approach to enhance spatial resolution. Nevertheless, aperture synthetic radiometers formed by such combinations usually consist of non-uniform antenna arrays (one-dimensional, two-dimensional, or three-dimensional). Compared with regular antenna arrays, they complicate the inversion of brightness temperature (BT) images. This paper proposes NASRT, a transformer-based inversion method designed for multi-dimensional non-uniform ASRs. The network extracts and fuses spectral and UVW spatial distribution features from the visibility function (VF), and introduces a learnable position weight matrix during training to capture spatial information of the non-uniform array. Through supervised learning, NASRT effectively maps the VF to BT images. Simulations across 1-D, 2-D, and 3-D NASR scenes demonstrate that NASRT achieves higher accuracy and stability than traditional methods. In a 1-D NASR indoor experiment, the proposed method also shows improved inversion accuracy and lower sidelobes, validating its effectiveness.}, archive = {J_ESWA}, author = {Jian Dong and Jiaxin Li and Chengwang Xiao and Rigeng Wu and Haofeng Dou and Wenjing Wang and Yuanchao Wu and Liangbing Chen}, doi = {10.1016/j.eswa.2025.129641}, journal = {Expert Systems with Applications}, month = {3}, pages = {129641}, shortjournal = {Expert Syst. Appl.}, title = {A transformer network for multi-dimensional nonuniform aperture synthesis radiometer image inversion}, volume = {298}, year = {2026}, }
• (2026). OceanAgent: A small-scale multi-modal assistant for ocean exploration. ESWA, 298, 129640. (www)

  @article{ , abstract = {Extraction of key information and the subsequent generation of actionable knowledge from multimodal data are critical for ocean exploration. Traditional knowledge generation methods rely heavily on expert experience and are labor-intensive. Recently, Large Multimodal Models (LMMs) have shown exceptional capabilities for knowledge generation from multimodal data in many complex tasks. These models also have potential to assist knowledge mining in ocean exploration. However, two major challenges faced by the LMMs when used in ocean exploration include the scarcity of ocean instruction-following data and the degradation of underwater visual environments. In this paper, a small-scale LMM for ocean exploration, named the OceanAgent, is designed. First, a swarm-intelligence-based leaderless multi-agent collaboration framework is proposed to generate visual instruction-following data. Subsequently, we present a visual-language connector to simultaneously extract multi-scale features. It is formed by integrating a multi-scale residual network with a multi-layer perceptron, which can enhance the model’s performance on severely low-quality images. Experiments show that the proposed method for constructing visual instruction-following datasets improves both the textual quality and visual dialogue. When severely degraded ocean visual data are processed using the trained OceanAgent, the image description accuracy and image comprehension are improved by 23.6 % and 21.6 %, respectively, compared to existing models. Additionally, the model demonstrates superior domain expertise, with a 95.7 % win rate in dialogue quality assessments.}, archive = {J_ESWA}, author = {Yun Xu and Yue Liu and Junpeng Shang and Jianmin Lin and Dongfang Ma}, doi = {10.1016/j.eswa.2025.129640}, journal = {Expert Systems with Applications}, month = {3}, pages = {129640}, shortjournal = {Expert Syst. Appl.}, title = {OceanAgent: A small-scale multi-modal assistant for ocean exploration}, volume = {298}, year = {2026}, }
• (2026). SPS-GAD: Spectral-spatial graph structure learning for anomaly detection in heterophilic graphs. ESWA, 298, 129639. (www)

  @article{ , abstract = {Graph anomaly detection(GAD) plays a critical role in fields such as fraud detection and network security. Although existing graph anomaly detection methods have achieved promising performance, most graph neural networks (GNNs) rely on the homophily assumption, which presumes that connected nodes share similar labels. However, real-world graphs frequently exhibit pronounced heterophily. Owing to class imbalance, normal nodes tend to have lower heterophily while anomalous nodes display higher heterophily. Furthermore, feature inconsistency induced by node camouflage exacerbates the detection challenge, rendering many existing approaches ineffective. To overcome these limitations, we propose SPS-GAD, a spectral-spatial graph structure learning framework specifically designed for detecting anomalous nodes in heterophilic graphs. First, to alleviate the feature inconsistency resulting from node camouflage, we develop a node reconstruction module that learns intermediate node representations to mitigate camouflage-induced bias, and applies spectral filters to extract the graph’s inherent structural features. Second, to address the heterophily disparities arising from class imbalance, we introduce a subgraph-type-aware spectral filtering module that leverages edge scores generated by an edge partitioner to segregate the graph into homophilic, ambiguous, and heterophilic subgraphs. Distinct spectral filters are subsequently applied to capture features across various frequency bands. Additionally, we integrate a neighbor-type-aware graph attention module that employs edge scores within an attention mechanism to guide the feature aggregation process, thereby enhancing spatial representation learning. Experimental evaluations on six real-world datasets reveal that SPS-GAD significantly outperforms all baseline methods in key metrics such as F1-Macro and AUC, thereby confirming its effectiveness in graph anomaly detection. The source code is publicly available at https://github.com/cozy24/SPS-GAD .}, archive = {J_ESWA}, author = {Chen Zhu and Yaying Zhang}, doi = {10.1016/j.eswa.2025.129639}, journal = {Expert Systems with Applications}, month = {3}, pages = {129639}, shortjournal = {Expert Syst. Appl.}, title = {SPS-GAD: Spectral-spatial graph structure learning for anomaly detection in heterophilic graphs}, volume = {298}, year = {2026}, }
• (2026). Pareto optimization of two-agent scheduling on parallel batch machines. ESWA, 298, 129637. (www)

  @article{ , abstract = {This paper considers a Pareto optimization problem of scheduling jobs of two competing agents on parallel batch machines. The jobs have equal processing times and non-identical job sizes. The objective is to find Pareto optimality and the corresponding schedules for minimizing both agents’ makespans. We analyze and identify an approximate Pareto region with a guarantee of 2-approximate Pareto optimal. We propose an integrated algorithm to find the approximate Pareto optimal points. Our computational study shows that the proposed algorithm outperforms the widely used non-dominated sorting genetic algorithm (NSGA-II), and that the obtained approximate Pareto optimal front is very close to the Pareto optimal front.}, archive = {J_ESWA}, author = {Cui-Lin Zhang and Guo-Qiang Fan}, doi = {10.1016/j.eswa.2025.129637}, journal = {Expert Systems with Applications}, month = {3}, pages = {129637}, shortjournal = {Expert Syst. Appl.}, title = {Pareto optimization of two-agent scheduling on parallel batch machines}, volume = {298}, year = {2026}, }
• (2026). Hybrid fleet composition and scheduling for road-based cross-border logistics under cost differentiation: A bi-level programming approach. ESWA, 298, 129636. (www)

  @article{ , abstract = {Under the dual pressure of explosive growth in cross-border e-commerce demand and increasing timeliness requirements from overseas customers, cross-border logistics service providers are compelled to establish logistics facilities and deploy fleets across multiple regions to ensure rapid response. However, during freight transportation, the lack of effective management over these complex and heterogeneous fleets—particularly in terms of fleet composition and routing decisions—has led to high transportation costs and low operational efficiency. This study is grounded in the practical operational context of cross-border logistics in the Guangdong–Hong Kong–Macau Greater Bay Area and models a multi-level, multi-node cross-border transportation network. To minimize the overall operational cost, the problem is addressed from two interrelated decision-making perspectives: fleet composition at the strategic level and routing planning at the operational level. Thus, a bi-level programming model is proposed to systematically capture the hierarchical structure and the logical relationship between these two decision layers. Furthermore, the model incorporates cost differences among trucks with different functional capabilities to reflect the significant disparity in logistics cost structures between domestic and overseas operations. To address the above multi-objective mixed-integer linear programming (MILP) problem, a tailored Non-dominated Sorting Genetic Algorithm II (MNSGA-II) is developed. Several key components of the algorithm are modified and enhanced to improve its search efficiency and solution quality in handling the problem’s complexity. Comparative experiments against classical algorithms demonstrate the superior solution quality and robustness of the proposed approach. The influence of cost differentials on composition and scheduling decisions is further analyzed, providing practical insights for the strategic planning of cross-border logistics systems.}, archive = {J_ESWA}, author = {Zhi Tang and Ting Qu and Yanghua Pan and George Q. Huang}, doi = {10.1016/j.eswa.2025.129636}, journal = {Expert Systems with Applications}, month = {3}, pages = {129636}, shortjournal = {Expert Syst. Appl.}, title = {Hybrid fleet composition and scheduling for road-based cross-border logistics under cost differentiation: A bi-level programming approach}, volume = {298}, year = {2026}, }
• (2026). A diversity-based niching differential evolution with neighborhood competition for nonlinear equation systems. ESWA, 298, 129635. (www)

  @article{ , abstract = {Solving nonlinear equation systems (NESs) has long been a fundamental challenge in the field of optimization. Due to the existence of multiple roots, such problems often exhibit complex and multimodal characteristics. Although numerous differential evolution-based algorithms have been developed to solve NESs, most of them employ only a single mutation operator, which is not adaptable to different problem scenarios. To this end, a diversity-based niching differential evolution with neighborhood competition (DNDE) is proposed to solve NESs. First, a control mechanism that takes into account population diversity and the evolutionary stage is proposed to adaptively assign appropriate mutation strategies to each subpopulation (niche), thereby enhancing the efficiency of root-finding. Second, a neighborhood priority competition mechanism is proposed to reduce cross-peak competition between populations, which ensures local convergence while improving global convergence. Finally, a reinitialization strategy based on opposition learning is introduced to guide the population toward more promising areas of the search space. Experimental results on 18 complex NESs and two real-world engineering problems show that DNDE outperforms many advanced algorithms in both root rate and success rate, demonstrating its effectiveness and value in practical applications.}, archive = {J_ESWA}, author = {Jianwei Li and Xinchao Zhao and Lingyu Wu and Yizhan Wu and Lingjuan Ye}, doi = {10.1016/j.eswa.2025.129635}, journal = {Expert Systems with Applications}, month = {3}, pages = {129635}, shortjournal = {Expert Syst. Appl.}, title = {A diversity-based niching differential evolution with neighborhood competition for nonlinear equation systems}, volume = {298}, year = {2026}, }
• (2026). TGNet: Texture-enhanced guidance network for RGB-D salient object detection. ESWA, 298, 129633. (www)

  @article{ , abstract = {RGB-D salient object detection achieves salient region localization in complex scenes by fusing RGB images and depth images. Existing methods typically employ two-stream networks to extract features separately followed by cross-modal fusion. However, differences between heterogeneous modalities can easily lead to feature degradation during cross-modal fusion, while the inherent noise interference in low-quality depth maps may generate cumulative effects during multi-stage propagation, severely constraining detection performance. To address these challenges, this paper proposes a texture-enhanced guided network. The core innovations lie in three aspects: during the feature encoding stage, a texture-enhanced module is constructed to utilize high-frequency texture information from RGB images through attention mechanisms for hierarchical optimization of depth features; in the feature fusion stage, a dual-path adaptive interaction module is designed to establish cross-modal semantic correlations via channel-spatial cooperative driving mechanisms, effectively suppressing redundant feature interference; for the decoding reconstruction stage, a dynamic hierarchical guidance mechanism is proposed to drive progressive calibration of low-level spatial details by high-level semantic features through learnable cross-scale transformation modules. Extensive experiments conducted on five benchmark datasets demonstrate that our method achieves competitive performance compared to other approaches.}, archive = {J_ESWA}, author = {Xiaogang Song and Hexiang Huang and Qin Zhao and Xinwei Guo and Xinhong Hei}, doi = {10.1016/j.eswa.2025.129633}, journal = {Expert Systems with Applications}, month = {3}, pages = {129633}, shortjournal = {Expert Syst. Appl.}, title = {TGNet: Texture-enhanced guidance network for RGB-D salient object detection}, volume = {298}, year = {2026}, }
• (2026). Dual-stream temporal-spatial forgery detection via phase-consistent edge features and 3D visual state-space modeling. ESWA, 298, 129632. (www)

  @article{ , abstract = {Deepfake, as a generative technology, has opened up new avenues for the development of the film, television, and art industries. However, its abusive use has triggered serious social security threats, such as infringement of portrait rights and the spread of misinformation, which has drawn widespread attention to research on deepfake detection techniques. Current deep learning-based face forgery detection methods face critical challenges: 1) insufficient focus on common forgery traces leads to poor generalization performance on datasets generated by unknown forgery methods; 2) traditional spatio-temporal feature fusion mechanisms struggle to balance the representational weights of spatial details and temporal dynamics, and exhibit inadequate robustness against post-processing operations like compression and cropping. To address these issues, this paper first designs a phase consistency edge artifact mining module is designed to extract common forgery traces from edge textures by leveraging the deep-phase information of images, significantly enhancing the model’s generalization ability. Second, a multi-frame synthesis strategy is designed to effectively integrate spatial and temporal features while balancing the network’s attention to these two feature domains. Third, a visual state-space model based on 3D scanning is designed, which for the first time employs the Mamba model to analyze spatio-temporal forgery patterns, notably improving the robustness of the model against unknown perturbations. Experimental results on standard benchmarks–FaceForensics++, Celeb-DFv2, WildDeepfake and DFDC(Preview)–demonstrate that the proposed method achieves state-of-the-art performance in three core dimensions: detection accuracy, cross-dataset generalization, and robustness against perturbations.}, archive = {J_ESWA}, author = {Zhong Chen and Siyang Wang and Zuxi Wang}, doi = {10.1016/j.eswa.2025.129632}, journal = {Expert Systems with Applications}, month = {3}, pages = {129632}, shortjournal = {Expert Syst. Appl.}, title = {Dual-stream temporal-spatial forgery detection via phase-consistent edge features and 3D visual state-space modeling}, volume = {298}, year = {2026}, }
• (2026). LSTT: Long short-term feature enhancement transformer for video small object detection. ESWA, 298, 129631. (www)

  @article{ , abstract = {Leveraging temporal information is crucial for small object detection in videos. Existing methods typically incorporate long-term or short-term temporal information uniformly, neglecting distinct cues from different frames that are essential for small object detection. In this paper, we propose LSTT, an end-to-end multi-frame fusion network that concurrently extracts global scene context from long-term frames and fine-grained appearance and motion cues from short-term frames. First, we introduce a progressive spatiotemporal sampling module that sparsely samples long-range frames and densely samples short-range frames. Second, we design a spatiotemporal alignment encoder module to extract frame-level temporal and spatial pixel features. Finally, We propose a long short-term feature aggregation module that employs a dynamic query generator to derive adaptive queries by implicitly modeling motion relationships among short-term frames, and guides a cascaded fusion of aggregated features from long-term, short-term, and current frames to fuse temporal information. Compared to state-of-the-art methods, our LSTT achieves absolute gains of 1.4 % and 2.1 % in detection precision on VisDrone-VID and UAVDT datasets, respectively.}, archive = {J_ESWA}, author = {Jinsheng Xiao and Wenbo Liu and Ruidi Chen and Yuchen Yan and Wei Yang}, doi = {10.1016/j.eswa.2025.129631}, journal = {Expert Systems with Applications}, month = {3}, pages = {129631}, shortjournal = {Expert Syst. Appl.}, title = {LSTT: Long short-term feature enhancement transformer for video small object detection}, volume = {298}, year = {2026}, }
• (2026). Scale-invariant information bottleneck for domain generalization. ESWA, 298, 129628. (www)

  @article{ , abstract = {One significant challenge in deep learning is the inability to effectively generalize to new data whose distribution differs from that of the training data. Hence, domain generalization has received increasing attention in related fields. Classical methods aim to identify an invariant predictor that can recognize invariant representations across all the training domains. However, these methods limit the model to rely solely on invariant representations, which hinders the learning of important finer details. To address this challenge, we propose a Scale-invariant Information Bottleneck (SIB) method to identify both invariant and scale-invariant features. We subsequently introduce a tractable loss function derived from the variational analysis. This novel method captures more detailed information, including fine textures and unique characteristics, while also eliminating irrelevant or spurious representations by using information bottleneck. Finally, extensive experiments conducted on Rotated MNIST, Colored MNIST, Colored Fashion-MNIST, PACS, Office-Home and Camelyon17-WILDS validate the effectiveness of our SIB method in addressing the domain generalization problems. Notably, our approach outperforms 14 existing methods with an average improvement of 4.74 %. More significantly, it surpasses 6 recent, related methods by an average of 2.21 %. Furthermore, we demonstrate the superiority of our method through the analysis of hidden feature maps and representations.}, archive = {J_ESWA}, author = {Mengyao Li and Jiangshe Zhang and Chunxia Zhang and Junmin Liu and Lizhen Ji}, doi = {10.1016/j.eswa.2025.129628}, journal = {Expert Systems with Applications}, month = {3}, pages = {129628}, shortjournal = {Expert Syst. Appl.}, title = {Scale-invariant information bottleneck for domain generalization}, volume = {298}, year = {2026}, }
• (2026). Optimal pricing for transfer of development rights under risk sharing: The context of china’s inter-provincial construction land quota trading. ESWA, 298, 129627. (www)

  @article{ , abstract = {Scientific and rational prices are determinant for the success of transfer of development rights (TDR). Nevertheless, previous studies largely overlook the multifaceted impacts of risks on pricing, hampering market participation and value revelation. This is especially relevant in the context of China’s inter-provincial construction land quota trading due to its broader scope and dynamic complexities. This study addresses this gap by proposing an integrated decision-making framework to identify TDR risk factors and determine the optimal pricing for TDR under risk sharing. Results show that among 28 identified risk factors across the trading lifecycle, pre-transaction (remediation application and remediation acceptance) risk factors exhibit lower weights (0.017 and 0.218) but demand greater responsibility from quota-sending governments (80.5% and 73.0%); quota transfer risk factors hold the highest weight (0.306), with nearly balanced responsibility sharing between trading parties; while post-transaction (remediation acceptance and post monitoring) risk factors (weighted at 0.215 and 0.218) should be borne mainly by quota-receiving governments (64.1% and 60.9%). A paradigmatic trading case study between Muli and Jiashan Counties empirically reveals that risk factors elevate the optimal price to 621171.89 yuan/mu—24.23% above the current national standard price—by increasing costs, reducing profits, decreasing the supply–demand ratio, and complicating ecological compensation. These findings underscore the importance of risk responsibility management and risk-based pricing mechanisms.}, archive = {J_ESWA}, author = {Jia-He Zhou and Yu-Jia Wei and Yu-Ming Zhu and Hong-Li Lin}, doi = {10.1016/j.eswa.2025.129627}, journal = {Expert Systems with Applications}, month = {3}, pages = {129627}, shortjournal = {Expert Syst. Appl.}, title = {Optimal pricing for transfer of development rights under risk sharing: The context of china’s inter-provincial construction land quota trading}, volume = {298}, year = {2026}, }
• (2026). Intelligent urban on-street parking space management for autonomous vehicles. ESWA, 298, 129626. (www)

  @article{ , abstract = {Curbside lanes are valuable spatial assets, with on-street parking, driving, and other travel modes competing for the space. Autonomous vehicle (AV) transport is expected to park at the curbside for diverse purposes, raising conflicts between driving and parking in the city centre. This study presents a framework to determine on-street parking configurations under different traffic flow and parking supply scenarios for the downtown region. The main contribution stems from solving the macro-level parking configuration problem using customised metaheuristics while considering microscopic AV operations. We tested the framework using a road network comprising a downtown central business district and adjacent urban areas. Among the considered metaheuristics, the discrete particle swarm optimisation outperformed the genetic algorithm in minimising network-level travel delays but at the cost of higher computational time. Three main empirical findings are derived. First , parking lanes are more likely to be assigned to edges in downtown areas or those with lower traffic and driving speeds. Second , high parking supply negatively affects the macroscopic fundamental diagram by increasing congestion and reducing flow efficiency, but such an effect diminishes in congested networks. Third , there exists an optimal parking supply level (40 % in the case study) for most flow rate conditions that can help reduce congestion. The proposed framework was validated through a case study in Midtown Manhattan, New York City. This study provides valuable insights for urban and transportation agencies to manage on-street parking lane assignments to balance parking and driving demands in the AV transport era. The approach has broader applicability as it is transferable to human-driven vehicles and mixed autonomy scenarios.}, archive = {J_ESWA}, author = {Qiming Ye and Prateek Bansal and Yuxiang Feng and Simon Hu and Panagiotis Angeloudis}, doi = {10.1016/j.eswa.2025.129626}, journal = {Expert Systems with Applications}, month = {3}, pages = {129626}, shortjournal = {Expert Syst. Appl.}, title = {Intelligent urban on-street parking space management for autonomous vehicles}, volume = {298}, year = {2026}, }
• (2026). Cross-session interest extraction for recommendation. ESWA, 298, 129625. (www)

  @article{ , abstract = {In recommender systems, due to device privacy restrictions, sometimes we can only obtain anonymous users’ interaction behavior within a single session. This type of recommendation is called session-based recommendation. Modeling users’ interest based on session data is one of the core issues in session-based recommendation. However, most existing methods only model users’ interest within individual sessions, neglecting information propagation across sessions. This paper addresses this challenge by designing a contrastive learning module based on clustering to model inter-session information propagation. Specifically, in addition to propagating information within sessions using hypergraph convolution, a cluster algorithm is applied to group all nodes across sessions. Then a contrastive learning loss is designed based on the clustering results to facilitate information propagation across sessions, thereby explicitly modeling the semantic similarity of similar items across different sessions. We call our model Clustering Hypergraph Neural Network (CluHNN). CluHNN explicitly learns the correlation between similar items across different sessions, improving the quality of item representations and, consequently, yielding better interest representations through cross-session information propagation. Experimental results on two real-world datasets show the effectiveness of the proposed CluHNN. For example, in terms of MRR@20, CluHNN achieves significant improvements of 1.3 % and 2.0 % relative gains over the strongest baseline, respectively.}, archive = {J_ESWA}, author = {Jin Jin and Chaoqun Li and Liangxiao Jiang}, doi = {10.1016/j.eswa.2025.129625}, journal = {Expert Systems with Applications}, month = {3}, pages = {129625}, shortjournal = {Expert Syst. Appl.}, title = {Cross-session interest extraction for recommendation}, volume = {298}, year = {2026}, }
• (2026). Incorporating estimated depth maps and multi-modal pretraining to improve salient object detection in optical remote sensing images. ESWA, 298, 129624. (www)

  @article{ , abstract = {As a burgeoning theme in optical remote sensing image (ORSI) analysis, salient object detection (SOD) plays a vital role in traffic monitoring, agriculture, disaster management, and other fields. However, the existing ORSI-SOD methods are all single-modal (RGB images primarily), which suffer from performance drop when facing complex scenes (e.g., intricate backgrounds, low contrast scenes, and similar objects). To address this challenge, we introduce estimated depth map to complement RGB image in ORSI-SOD for the first time, which provides 3D geometric cues to improve detection accuracy in complex scenes, thus advancing ORSI-SOD from single-modal to multi-modal. Furthermore, we design a novel pretraining framework: multi-modal reconstructed image pretraining (MMRIP) to pretrain SOD model in multi-modal ORSI-SOD. MMRIP initially utilizes a masked autoencoder (MAE) to restore the masked RGB image; subsequently, it feeds the restored RGB image and clean depth map to the SOD model to generate the saliency map, which can help SOD model more effectively integrate cross modal information and extract better feature. Besides, we present a simple RGB-D SOD model, namely SimSOD, which is pretrained by MMRIP for ORSI-SOD. SimSOD has two major components: DFormer (encoder) and MLP head (decoder). Specifically, we first input RGB image and depth data into the encoder to generate four multi-scale features, then use the decoder to fuse these features and yield the prediction result. Without bells and whistles, our proposed method outperforms the state-of-the-art methods on three public ORSI-SOD datasets. The code can be accessed at: https://github.com/Voruarn/MMRIP .}, archive = {J_ESWA}, author = {Yuxiang Fu and Wei Fang}, doi = {10.1016/j.eswa.2025.129624}, journal = {Expert Systems with Applications}, month = {3}, pages = {129624}, shortjournal = {Expert Syst. Appl.}, title = {Incorporating estimated depth maps and multi-modal pretraining to improve salient object detection in optical remote sensing images}, volume = {298}, year = {2026}, }
• (2026). An approach for linking dynamic network information models based on ontology matching. ESWA, 298, 129622. (www)

  @article{ , abstract = {Dynamic network information models are typically heterogeneous and isolated systems that impede effective interoperability, significantly hindering end-to-end service integration and data sharing across network segments. To address this challenge, we propose a new approach for linking heterogeneous dynamic network models based on ontology matching, which can be applied in various domains utilizing dynamic networks. For ontologies matching we use different existing duplicate detection algorithms but we reduce the computational complexity of ontology matching due to splitting initial set of matched entities into a number of subsets using domain knowledge. Using telecommunications as case study, we represent operator networks as knowledge graphs and match them with standardized model ontologies using business process context to create an Extended Operator Network Ontology. Our approach ensures linking of dynamic network models used in operators information systems that is of primary importance for implementing complex business processes, and providing integrated services while maintaining existing models.}, archive = {J_ESWA}, author = {Tianxing Man and Igor Kulikov and Jiafeng Yang and Nataly Zhukova}, doi = {10.1016/j.eswa.2025.129622}, journal = {Expert Systems with Applications}, month = {3}, pages = {129622}, shortjournal = {Expert Syst. Appl.}, title = {An approach for linking dynamic network information models based on ontology matching}, volume = {298}, year = {2026}, }
• (2026). An improved differential evolution algorithm combined with vector NFP and mixed-integer programming for solving 2D irregular layout problem. ESWA, 298, 129621. (www)

  @article{ , abstract = {The two-dimensional irregular layout problem, which involves placing convex or non-convex components within a confined boundary without overlaps, is NP-complete and widely encountered in industrial applications such as glass cutting, garment manufacturing, and packaging. To overcome the limitations of existing methods in computational efficiency and material utilization, we propose a new hybrid algorithm IDE-V-NFP-MIP: (1) An improved differential evolution (IDE) algorithm combines the memory mechanism to guide the crossover and mutation operations; (2) A vector No-Fit Polygon (V-NFP) algorithm effectively handles complex geometric constraints, including voids and degradation; (3) A mixed-integer programming (MIP) model ensures accurate layout and non-overlapping constraints. Experimental results demonstrate superior performance: IDE ranked first in CEC2022 Friedman tests, while practical applications show 22.10% reduction in board length and 41.47% improvement in filling rate. The framework successfully handles real-world garment cutting applications and large-scale problems up to 1,280 polygons, demonstrating significant improvements in both computational efficiency and solution quality for industrial layout optimization. The source code for the algorithm is available at https://github.com/xhj-6/IDE-V-NFP-MIP .}, archive = {J_ESWA}, author = {Huijie Xu and Qifang Luo and Yongquan Zhou}, doi = {10.1016/j.eswa.2025.129621}, journal = {Expert Systems with Applications}, month = {3}, pages = {129621}, shortjournal = {Expert Syst. Appl.}, title = {An improved differential evolution algorithm combined with vector NFP and mixed-integer programming for solving 2D irregular layout problem}, volume = {298}, year = {2026}, }
• (2026). Intelligent localization of FDIA in smart grids: A multi-level wavelet spatio-temporal graph embedding approach. ESWA, 298, 129620. (www)

  @article{ , abstract = {In the realm of smart grid security, the precise identification of False Data Injection Attack (FDIA) is crucial for ensuring the stable operation of power systems. Existing approaches for handling measurement data often overlook the correlation between local time–frequency variations caused by FDIA nodes and global spatial information in analyzing measurement data, leading to inaccurate localization. To address this issue, we propose a novel approach: a multilevel wavelet spatio-temporal map embedded FDIA localization method. Initially, a multi-resolution time–frequency signal decomposition model is utilized to separate the time–frequency mutation signals induced by FDIA from the measurement data using fast wavelet transform. Subsequently, a multi-channel time–frequency feature extraction technique is developed to capture the mutation characteristics of FDIA in time–frequency signals. This involves extracting detailed features of the time–frequency signals pre and post-attack via a multi-channel convolution operation encompassing “temporal-local-global” aspects. Finally, we propose an FDIA localization model based on multi-level graph wavelet embedding. The model embeds spatio-temporal information into time–frequency features via graph wavelet convolution and builds a spatio-temporal dependency map through multi-level neighborhood sampling. To mitigate measurement loss and noise, graph smoothing regularization and graph dropout are introduced during training. A graph attention mechanism further captures spatio-temporal dependencies among nodes, enabling accurate FDIA localization. Experimental results verify the effectiveness of the proposed method.}, archive = {J_ESWA}, author = {Zhaoyang Qu and Feng Liang and Nan Qu and Tao Jiang and Xiaoyu Xu}, doi = {10.1016/j.eswa.2025.129620}, journal = {Expert Systems with Applications}, month = {3}, pages = {129620}, shortjournal = {Expert Syst. Appl.}, title = {Intelligent localization of FDIA in smart grids: A multi-level wavelet spatio-temporal graph embedding approach}, volume = {298}, year = {2026}, }
• (2026). AI-driven 5G-IoT optimization: Q-learning for real-time energy and network resource management. ESWA, 298, 129619. (www)

  @article{ , abstract = {In recent times, the integration of the 5G-enabled Internet of Things has revolutionized through high-speed data transmission, ultra-low latency, and interconnectivity of massive devices. However, the proliferation of 5G-enabled Internet of Things introduces major challenges, such as energy inefficiency and unreliable data delivery in the resource constrained Internet of Things devices. This research proposes a novel Q-Learning-based optimization framework tailored to address these challenges by integrating Radio Frequency energy harvesting, adaptive beamforming, and dynamic resource allocation within the massive Multiple-Input-Multiple-Output system. The proposed model utilizes reinforcement learning to manage the network resources including modulation schemes, beamforming, and energy allocation. By modeling the optimization problem as a Markov Decision Process, the proposed framework dynamically adapts to real-time network conditions to enhance energy efficiency, reliable data delivery, and throughput. The experimental validation demonstrates that the Q-Learning-based strategy effectively optimizes the energy efficiency as well as data transmission and achieves a higher energy efficiency of 98.87 %, higher packet delivery ratio of 98.85 %, lower latency of 1.5 ms, and higher throughput of 200Mbps compared to existing methodologies. This result indicates that the proposed Q-Learning-based framework has the potential to enhance the sustainability and reliability of the 5G-enabled Internet of Things.}, archive = {J_ESWA}, author = {Bavethra Murthy and Palani Uthirapathy}, doi = {10.1016/j.eswa.2025.129619}, journal = {Expert Systems with Applications}, month = {3}, pages = {129619}, shortjournal = {Expert Syst. Appl.}, title = {AI-driven 5G-IoT optimization: Q-learning for real-time energy and network resource management}, volume = {298}, year = {2026}, }
• (2026). Multi-behavioral recommendation algorithm based on decoupled graph convolution. ESWA, 298, 129618. (www)

  @article{ , abstract = {Traditional recommendation models primarily rely on display feedback and typically utilize a single type of user-item interaction data, which often results in significant data sparsity issues. In contrast, multi-behavioral recommendation models leverage various behaviors such as browsing, favoriting, and other interactions. These additional behaviors help improve the prediction of user-item interactions. Existing multi-behavioral recommendation methods often overlook the potential factors influencing multi-behavioral interactions and the differences between various behavior types. In this study, we introduce a multi-behavioral recommendation algorithm utilizing decoupled graph convolution (MBR-DGC), which effectively mitigates the data sparsity of the target behaviors and improves recommender system performance by capturing the differences between the semantics of different behaviors. Specifically, we construct multiple non-overlapping independent isomorphic graphs and separate potential factors affecting the interactions among users, items, and behaviors using decoupled convolutional networks to reconstruct the node features of users in different behaviors. Afterwards, multi-behavioral features of users are aggregated using contrastive learning to achieve personalized multi-behavioral information aggregation. Experimental results on multiple datasets show that MBR-DGC effectively leverages multi-behavioral data, significantly enhancing recommendation performance compared to other state-of-the-art methods.}, archive = {J_ESWA}, author = {Xu Yu and Pengju Ding and Jie Yu and Junyu Lin and Lei Guo and Guanfeng Liu and Liang Xi}, doi = {10.1016/j.eswa.2025.129618}, journal = {Expert Systems with Applications}, month = {3}, pages = {129618}, shortjournal = {Expert Syst. Appl.}, title = {Multi-behavioral recommendation algorithm based on decoupled graph convolution}, volume = {298}, year = {2026}, }
• (2026). An information aggregation decision making method for UAV swarm intelligence system based on joint communication and proximal strategy. ESWA, 298, 129617. (www)

  @article{ , abstract = {The swarm intelligence aggregation system represents a key capability in current-generation UAV swarm, demonstrating robust collective intelligence. Currently, leveraging Multi-Agent Deep Reinforcement Learning (MADRL) offers a promising approach for building UAV swarm intelligence aggregation systems. However, the MADRL methods are difficult to cope with the challenge of exponential increase in computation when facing the collaboration problem of large-scale swarms, and the agents also have the problem of partial observability of the environment. This paper proposes an Information Aggregation Decision Method for UAV swarm based on Joint Communication and Proximal Strategy (IADM-JCPS). This method designs a communication information aggregation (CIA) network to enable UAVs to gather observation information from neighbor UAVs, and uses the attention mechanism to screen important information. Then, the aggregated information is used as part of the input of the policy network to increase the information diversity of the decision-making process. Finally, the gradient clipping mechanism is used to trim the policy gradient to enhance the stability of the training process. A UAV swarm multi-target tracking (MTT) mission scenario is designed to verify the effectiveness of the proposed IADM-JCPS algorithm. Experimental results show that the proposed algorithm is superior to the baseline algorithm in terms of task collaboration and scalability.}, archive = {J_ESWA}, author = {Zhaotian Wei and Ruixuan Wei}, doi = {10.1016/j.eswa.2025.129617}, journal = {Expert Systems with Applications}, month = {3}, pages = {129617}, shortjournal = {Expert Syst. Appl.}, title = {An information aggregation decision making method for UAV swarm intelligence system based on joint communication and proximal strategy}, volume = {298}, year = {2026}, }
• (2026). Learning-based trajectory planning for AGVs in dynamic environment. ESWA, 298, 129616. (www)

  @article{ , abstract = {In this work, we present a learning-based framework for rapid trajectory planning of autonomous ground vehicles (AGVs) in dynamic environments. The approach integrates optimization techniques with deep learning to design a real-time planner capable of generating kinematically feasible trajectories. A continuous iterative method is first developed for dataset construction, enabling efficient generation of optimal trajectory sets. Based on this dataset, a neural network is trained to learn the mapping between AGV states and actions while capturing their temporal dependencies. During online planning, the trained model produces decision actions from the current state and sensor feedback, enabling real-time planning of safe and feasible trajectories. Results demonstrate the effectiveness of the proposed framework.}, archive = {J_ESWA}, author = {Runda Zhang and Zhida Xing and Senchun Chai and Yuanqing Xia and Runqi Chai}, doi = {10.1016/j.eswa.2025.129616}, journal = {Expert Systems with Applications}, month = {3}, pages = {129616}, shortjournal = {Expert Syst. Appl.}, title = {Learning-based trajectory planning for AGVs in dynamic environment}, volume = {298}, year = {2026}, }
• (2026). Model-agnostic post-hoc explainability for recommender systems. ESWA, 298, 129608. (www)

  @article{ , abstract = {Recommender systems often benefit from complex feature embeddings and deep learning algorithms, which deliver sophisticated recommendations that enhance user experience, engagement, and revenue. However, these methods frequently reduce the interpretability and transparency of the system. In this research, we develop a systematic application, adaptation, and evaluation of deletion diagnostics in the recommender setting. The method compares the performance of a model to that of a similar model trained without a specific user or item, allowing us to quantify how that observation influences the recommender, either positively or negatively. To demonstrate its model-agnostic nature, the proposal is applied to both Neural Collaborative Filtering (NCF), a widely used deep learning-based recommender, and Singular Value Decomposition (SVD), a classical collaborative filtering technique. Experiments on the MovieLens and Amazon Reviews datasets provide insights into model behavior and highlight the generality of the approach across different recommendation paradigms.}, archive = {J_ESWA}, author = {Irina Arévalo and Jose L. Salmeron}, doi = {10.1016/j.eswa.2025.129608}, journal = {Expert Systems with Applications}, month = {3}, pages = {129608}, shortjournal = {Expert Syst. Appl.}, title = {Model-agnostic post-hoc explainability for recommender systems}, volume = {298}, year = {2026}, }
• (2026). AERO-net: AutoEncoder-driven residual optimization network for latent bias correction of WRF-ROMS. ESWA, 298, 129607. (www)

  @article{ , abstract = {Reliable assessment of precipitation is crucial for incorporating meteorological and hydrological research into industrial and agricultural applications. Accurately estimating precipitation is a challenging task. In addressing this problem, we propose to develop AERO-Net, a novel deep learning framework designed to correct spatial, temporal, and amplitude biases in WRF-ROMS precipitation data. The integration of the Weather Research and Forecasting (WRF) model with the Regional Ocean Modeling System (ROMS) makes it a valuable tool for precipitation forecasting. AERO-Net incorporates autoencoders (AEs) for handling fluctuation and generalizing latent space representations, a latent module (LM) for transforming WRF-ROMS data into bias-corrected representations, a residual module (RM) for error minimization via boost, and a calibration module (CM) for improving near-zero precipitation. Empirical results show that AERO-Net achieves a balanced error reduction across precipitation cohorts grouped by intensity, reducing the macro-averaged root mean square error (macro RMSE) by 3.6 mm/day and the macro-averaged mean absolute deviation (macro MAD) by 0.68 mm/day compared to the original WRF-ROMS. AERO-Net is seen to improve the correlation coefficient (CC) by 26.32 %, increasing it from 0.38 to 0.48, in comparison to the original WRF-ROMS. These findings underscore its potential as an effective solution for enhancing precipitation estimates in high-resolution modeling systems.}, archive = {J_ESWA}, author = {Passin Pornvoraphat and Kanoksri Sarinnapakorn and Ken-Ichi Fukui and Peerapon Vateekul}, doi = {10.1016/j.eswa.2025.129607}, journal = {Expert Systems with Applications}, month = {3}, pages = {129607}, shortjournal = {Expert Syst. Appl.}, title = {AERO-net: AutoEncoder-driven residual optimization network for latent bias correction of WRF-ROMS}, volume = {298}, year = {2026}, }
• (2026). Contrastive learning and physics oriented evaluation for advanced segmentation in electron tomography. ESWA, 298, 129606. (www)

  @article{ , abstract = {Deep learning methods are now achieving strong results for segmentation tasks, and the standard metric for evaluating methods is the Intersection over Union (IOU). However, we show in this paper that IOU is not efficient in evaluating the quality of segmentation for electron tomography (ET) images of zeolites. We perform a physics-oriented evaluation to ensure that the segmentation results yield coherent physical measures. We also formalize Mixed Supervised / Self-Supervised Contrastive Learning Segmentation (M3S-CLS), a semi-supervised approach using a contrastive learning approach that uses expert annotations to train the neural network model. A detailed comparison of this method with a standard cross-entropy-based model is provided. In addition, we publish a database of five fully segmented ET volumes along with corresponding baseline results. The code and the database is available at http://gitlab.univ-st-etienne.fr/labhc-iscv/M3S-CLS .}, archive = {J_ESWA}, author = {Cyril Li and Christophe Ducottet and Maxime Moreaud and Sylvain Desroziers and Valentina Girelli Consolaro and Virgile Rouchon and Ovidiu Ersen}, doi = {10.1016/j.eswa.2025.129606}, journal = {Expert Systems with Applications}, month = {3}, pages = {129606}, shortjournal = {Expert Syst. Appl.}, title = {Contrastive learning and physics oriented evaluation for advanced segmentation in electron tomography}, volume = {298}, year = {2026}, }
• (2026). DSANet: Deep surrounding-aware network for camouflaged object detection via cross-refinement mirror strategy. ESWA, 298, 129605. (www)

  @article{ , abstract = {Camouflaged objects often closely resemble their surroundings, causing standard RGB images to be confounded by background, texture, and color variations. This often leads to incomplete or absent target segmentation, reducing overall accuracy. To address this issue, we present a Deep Surrounding-Awareness Mirror Network (DSANet) for camouflaged object detection, leveraging depth information to expose objects incongruent with their environment, thus improving localization accuracy. First, a Convolutional Spatial Gating module processes batched RGB and depth inputs, suppressing extraneous background noise while isolating fine-grained segmentation and structural features and unifying channel representation. Subsequently, a Deep Surrounding-Awareness Localization module and a Contour-Guided Integrity Aggregation module collaboratively refine and merge multi-level features, focusing on the global form of camouflaged objects while iteratively enhancing segmentation detail. Finally, a Guided Residual Channel Attention module further refines low-layer structural cues. Extensive experiments on ten challenging benchmark datasets using four widely used evaluation metrics demonstrate that our method exhibited superior performance, outperforming 40 state-of-the-art methods. The results demonstrate the versatility of our model. The source code and results of our method are available at https://github.com/lixu11/DSANet.}, archive = {J_ESWA}, author = {Xu Li and Xiaosheng Yu and Peng Chen}, doi = {10.1016/j.eswa.2025.129605}, journal = {Expert Systems with Applications}, month = {3}, pages = {129605}, shortjournal = {Expert Syst. Appl.}, title = {DSANet: Deep surrounding-aware network for camouflaged object detection via cross-refinement mirror strategy}, volume = {298}, year = {2026}, }
• (2026). FTUAttack: Feature truncation unrestricted attack based on stable diffusion model. ESWA, 298, 129604. (www)

  @article{ , abstract = {In the field of adversarial example generation and defense, compared to restricted attacks with L p -norm constraints, unrestricted attacks without L p -norm constraints emanate better visual imperceptibility. Existing unrestricted attacks typically manipulate the semantic content of examples (e.g. texture or color) to generate adversarial examples. However, current works usually ignore multifaceted features or loss optimization strategy, which limits attack performance. In this paper, we draw inspiration from stable diffusion model and propose a unrestricted attack method called Feature Truncation Unrestricted Attack (FTUAttack) to achieve both better transferability and imperceptibility. Specifically, we promote the performance of unrestricted attacks from the perspectives of both diffusion principle and feature truncation for the first time. Firstly, we propose a Global Deep Feature Extractor (GDFE) module to truncate global feature for the subsequent diffusion denoising process. Secondly, to further boost the transferability, we design a novel Critical Latent Feature Extractor (CLFE) module to obtain critical local feature that need to be truncated during the denoising process and investigate the influence of the different segmentation ways on critical local feature. Thirdly, we propose Multi-Loss Fusion (MLF) strategy to balance the conflict between perturbations and examples’ quality by guiding the optimization direction. Extensive experiments on various model structures and datasets demonstrate the superiority of our attack over the existing attack methods.}, archive = {J_ESWA}, author = {Shaojie Han and Gangzheng Zhai and Kun Chen and Shihui Zhang}, doi = {10.1016/j.eswa.2025.129604}, journal = {Expert Systems with Applications}, month = {3}, pages = {129604}, shortjournal = {Expert Syst. Appl.}, title = {FTUAttack: Feature truncation unrestricted attack based on stable diffusion model}, volume = {298}, year = {2026}, }
• (2026). Spatial-temporal hierarchical decoupled masked autoencoder: A self-supervised learning framework for electrocardiogram. ESWA, 298, 129603. (www)

  @article{ , abstract = {The difficulty of labeling Electrocardiogram (ECG) has prompted researchers to use self-supervised learning to enhance diagnostic performance. Masked autoencoders (MAE) are a mainstream paradigm where models learn a latent representation of the signal by reconstructing masked portions of the ECG. However, existing methods lack a specific design for the spatial–temporal characteristics of ECG. Specifically, leads represent spatial projections of cardiac activity, while timestamps capture temporal patterns, and the two correspond to different axes of information. Existing MAE frameworks tend to unify them prematurely, potentially weakening critical local dependencies. In this paper, we propose a Spatial-Temporal Hierarchical Decoupled Masked Autoencoder (STHD-MAE). This framework decouples ECG into isolated leads or time steps in the shallow layer to capture local dependencies with different views, then aligns spatial–temporal representations and re-establishes global dependencies in the deep layer to comprehensively represent pathological information. We also design a medical report fusion module during pre-training, which uses cross-attention to align the ECG report text encoded by a medical language model with the signal’s latent representation, thereby guiding the encoder to focus on pathological information through implicit cross-modal learning. We validate the effectiveness of STHD-MAE on multiple downstream classification and reconstruction tasks. The results show that STHD-MAE outperforms existing self-supervised learning methods by approximately 2% in F1-scores for both coarse-grained and fine-grained classification performance, and its reconstruction quality also exceeds the baseline generative model.}, archive = {J_ESWA}, author = {Xiaoyang Wei and Zhiyuan Li and Yuanyuan Tian and Mengxiao Wang and Yanrui Jin and Weiping Ding and Chengliang Liu}, doi = {10.1016/j.eswa.2025.129603}, journal = {Expert Systems with Applications}, month = {3}, pages = {129603}, shortjournal = {Expert Syst. Appl.}, title = {Spatial-temporal hierarchical decoupled masked autoencoder: A self-supervised learning framework for electrocardiogram}, volume = {298}, year = {2026}, }
• (2026). GARE-net: Geometric contextual aggregation and regional contextual enhancement network for image-text matching. ESWA, 298, 129602. (www)

  @article{ , abstract = {The local correspondence learning has gained increasing attention in image-text matching, which establishes fine-grained alignments between image regions and textual words to improve both interpretability and accuracy. While these approaches have made significant progress in identifying meaningful semantic correspondences, one critical limitation persists in current methods, i.e., overlooking the crucial spatial position information of visual regions in cross-modal interaction. To address this challenge, we propose a novel Geometric contextual Aggregation and Regional contextual Enhancement Network (GARE-Net) that introduces two innovative components: the Geometric Contextual Feature Aggregation (GCFA) module and the Regional Contextual Feature Enhancement (RCFE) module. Specifically, GCFA generates the spatial geometric information of visual regions to enhance the region features by feature aggregation. RCFE further refines the aggregated region features by constructing a region graph and graph convolution. Extensive experiments and analyses are conducted on Flickr30k and MSCOCO to evaluate the importance of our framework. The results demonstrate the superiority of our method in image-text matching. Moreover, the ablation studies and visualization case studies also highlight the importance of geometric contextual feature aggregation and regional contextual feature enhancement. The code is available at https://github.com/chinaBoy123/GARE-Net .}, archive = {J_ESWA}, author = {Fangming Zhong and Tao Zhou and Zhikui Chen and Suhua Zhang}, doi = {10.1016/j.eswa.2025.129602}, journal = {Expert Systems with Applications}, month = {3}, pages = {129602}, shortjournal = {Expert Syst. Appl.}, title = {GARE-net: Geometric contextual aggregation and regional contextual enhancement network for image-text matching}, volume = {298}, year = {2026}, }
• (2026). Quantum probabilistic group consensus decision-making model based on matrix fluctuation grey correlation for engineering bid evaluation. ESWA, 298, 129600. (www)

  @article{ , abstract = {The existing quantum group decision-making models face significant challenges in the bid evaluation of engineering projects, including the strong subjectivity of expert evaluations, the difficulty in aggregating expert opinions, the large gap of expert opinions, and the complexity of expert psychological behaviors. To address these issues, this paper proposes a novel quantum probabilistic group consensus decision-making model based on matrix fluctuation grey correlation. Firstly, a quantum Bayesian network is constructed to aggregate expert opinions and capture the interference effect among experts. Secondly, the matrix fluctuation grey correlation degree is defined and applied to the calculation of quantum interaction terms that reflect the intricate psychological behavior of experts. Subsequently, a decision item search model is proposed and applied to adjust preferences during the consensus reaching process, thereby narrowing the gap of expert opinions. The consensus effect optimization model is utilized to determine optimal values for unknown parameters within this process, effectively reducing the subjectivity of expert evaluations. Finally, the proposed model is applied to a bid evaluation of bridge anti-collision engineering project, which verifies the feasibility and effectiveness of the model, and evaluates the stability and superiority of the model through sensitivity analysis and comparative analysis.}, archive = {J_ESWA}, author = {Jiuru Zhu and Xinping Xiao and Congjun Rao}, doi = {10.1016/j.eswa.2025.129600}, journal = {Expert Systems with Applications}, month = {3}, pages = {129600}, shortjournal = {Expert Syst. Appl.}, title = {Quantum probabilistic group consensus decision-making model based on matrix fluctuation grey correlation for engineering bid evaluation}, volume = {298}, year = {2026}, }
• (2026). DPCND: A dual-population based evolutionary approach for critical node detection problem in complex networks. ESWA, 298, 129599. (www)

  @article{ , abstract = {Critical node detection is an important tool for measuring network robustness. The main purpose of critical node detection is to detect a set of nodes that cause the greatest damage to the network connectivity, and it has been applied in many fields such as social network analysis and traffic network management. As a classic non-deterministic polynomial time complete problem, critical node detection faces enormous challenges with the continuous expansion of network size. The existing methods are difficult to achieve a good balance between effectiveness and efficiency, especially when the scale of complex networks becomes larger. To this end, this paper proposes a dual population based critical node detection method (DPCND) to effectively and efficiently obtain a set of critical nodes, which utilizes the co-evolution of auxiliary population generated from reduced graph and main population generated from original graph to find the optimal solution. In the proposed algorithm, a dual population interaction mechanism consists of influence and expansion strategies is proposed for information exchange, where the influence strategy transfers candidate good solutions from the auxiliary population to the main population to improve search efficiency, and the expansion strategy provides node information of the main population to guide the expansion of search space for the auxiliary population. Finally, the experimental results on 20 real-world complex networks clearly demonstrate the effectiveness of the proposed algorithm comparing to the state-of-the-arts.}, archive = {J_ESWA}, author = {Lei Zhang and Xinyi Feng and Yuanyuan Ge and Zhanpeng Wang and Haipeng Yang}, doi = {10.1016/j.eswa.2025.129599}, journal = {Expert Systems with Applications}, month = {3}, pages = {129599}, shortjournal = {Expert Syst. Appl.}, title = {DPCND: A dual-population based evolutionary approach for critical node detection problem in complex networks}, volume = {298}, year = {2026}, }
• (2026). A two-stage evolutionary algorithm with restart scheme for an integrated robot-task-scheduling and vehicle-dispatch-scheduling problem. ESWA, 298, 129598. (www)

  @article{ , abstract = {In modern agriculture, efficiently picking and distributing fresh product is crucial for the competitiveness of smart farms within the globalized agricultural market. However, the integrated scheduling problem involving both picking and distribution processes has received limited attention in existing research. To bridge this gap, this study establishes a mathematical model with dual objectives: (1) minimizing the picking completion time and (2) reducing penalties incurred due to early or delayed deliveries. A novel two-stage evolutionary algorithm incorporating a restart mechanism is proposed to effectively balance the optimization of these objectives with a high degree of consistency. The algorithm features an efficient encoding scheme and advanced genetic operators, specifically designed to enhance exploration and exploitation based on the characteristics of the problem. A comprehensive set of test instances is generated and the proposed method is benchmarked against several state-of-the-art metaheuristics from the literature. Experimental results demonstrate that the proposed algorithm outperforms the competing approaches by a significant margin for solving the problem under consideration.}, archive = {J_ESWA}, author = {Yiran Pan and Xuan He and Nan Li and Zhonghua Miao}, doi = {10.1016/j.eswa.2025.129598}, journal = {Expert Systems with Applications}, month = {3}, pages = {129598}, shortjournal = {Expert Syst. Appl.}, title = {A two-stage evolutionary algorithm with restart scheme for an integrated robot-task-scheduling and vehicle-dispatch-scheduling problem}, volume = {298}, year = {2026}, }
• (2026). An initial value insensitive method for phase equilibrium calculation: Constrained quadratic interpolation optimization algorithm. ESWA, 298, 129597. (www)

  @article{ , abstract = {The successful simulation of constrained differential evolution (CDE) algorithm for solving phase equilibrium calculation has first verified that heuristic optimization algorithms are effective ways to solve this kind of problems. Their insensitivity to initial values overcomes the limitations associated with two kinds of traditional methods, i.e., direct solution methods based on Newton’s method and indirect solution methods based on thermodynamic principles. This article proposes a constrained quadratic interpolation optimization algorithm (CQIO) for obtaining the satisfactory solutions of phase equilibrium calculation under given volume, temperature, and moles (NVT-flash). The proposed CQIO regards the total Helmholtz free energy of a NVT-flash problem as its objective function, while the moles vector and volume of a certain phase as its decision variables. The consistency between the four cases’ experimental results of CQIO and those of published articles demonstrates the effectiveness of CQIO in solving NVT-flash problems. Then the computational overhead and algorithmic stability of CQIO were analyzed. In Cases 1, 2 and 3, the average CPU time of CQIO compared to CDE has increased by 46.98 % , 54.56 % and 21.02 % respectively. The Std values of CQIO are significantly smaller than those of CDE in all cases except for Example 2. The proposed CQIO greatly promotes the application of heuristic algorithm in the field of phase equilibrium calculation.}, archive = {J_ESWA}, author = {Wangyu Tong and Baoduo Su and Yaqian Zhan}, doi = {10.1016/j.eswa.2025.129597}, journal = {Expert Systems with Applications}, month = {3}, pages = {129597}, shortjournal = {Expert Syst. Appl.}, title = {An initial value insensitive method for phase equilibrium calculation: Constrained quadratic interpolation optimization algorithm}, volume = {298}, year = {2026}, }
• (2026). Knowledge graph life cycle for cognitive agents – A case study on automated negotiation in smart grids. ESWA, 298, 129596. (www)

  @article{ , abstract = {Knowledge Graphs (KGs) can enhance cognitive artificial agents by improving their semantic understanding, adaptability to dynamic contexts, explainability, continuous learning, and informed, case-based decision-making. However, the development of any KG rarely follows an explicit and structured procedure to improve its consistency, modifiability, and interoperability. This ultimately restricts the utility of KGs for real-world agentic AI systems. To overcome this limitation, this paper proposes a life cycle for developing and evolving KGs in domain-specific applications, encompassing three phases to i) conceptualize the problem domain and competency questions, ii) formalize a schema using ontologies, and iii) implement the KG to foster agents’ cognition through queries and graph data science. The proposed approach is validated with a case study on context-aware automated negotiations within smart grids, where agents negotiate for energy trading while considering private and contextual circumstances. Agents may take actions with or without the aid of a KG, and can adopt one of three negotiation strategy configurations: heuristic, metaheuristic, or reinforcement learning-based. Negotiation outcomes consistently indicate that, regardless of the configuration employed, the use of a KG improves agents’ rewards by at least 1.21 % and up to 90.91 %. Results highlight that the proposed life cycle enables the integration of contextual and domain-specific data and metadata into a KG that enhances agents’ learning, while also allowing for the development and selection of relevant queries and graph data science algorithms to improve the strategic behavior of negotiation agents with cognitive abilities.}, archive = {J_ESWA}, author = {Dan E. Kröhling and Ernesto C. Martínez}, doi = {10.1016/j.eswa.2025.129596}, journal = {Expert Systems with Applications}, month = {3}, pages = {129596}, shortjournal = {Expert Syst. Appl.}, title = {Knowledge graph life cycle for cognitive agents – A case study on automated negotiation in smart grids}, volume = {298}, year = {2026}, }
• (2026). Research on multi-objective optimization of multi-endpoint VRP with time window for the distribution of seasonal products by multi-homing heterogeneous fleets. ESWA, 298, 129595. (www)

  @article{ , abstract = {This paper addresses a novel VRP variant integrating seasonal demand fluctuations, heterogeneous vehicle sources, and multi-endpoint constraints, focusing on the distribution of seasonal products in a steel parts enterprise. It tackles the complex vehicle routing problem with time windows involving heterogeneous fleets, which encompass different vehicle sources (owned and rented), types (fuel-powered and electric), capacities, ranges, and endpoints. To balance enterprise profitability, greenhouse gas emissions, and environmental quality, we develop a mathematical model centered on optimizing distribution costs, greenhouse gas emissions, and vehicle utilization. Drawing inspiration from ancient competitive activities, we propose a novel Huashan Swords Algorithm (HSSA). Through simulations using real enterprise data, we demonstrate the HSSA’s effectiveness, with comparative experiments against existing advanced algorithms highlighting its superiority. Applying the algorithm to design logistics distribution schemes, we conduct in-depth tests considering different customer groups and fuel station distributions. Analyzing the results from the perspectives of profitability, emissions, and environmental quality, we offer targeted operational suggestions for the enterprise based on its situation, geographical characteristics, and fiscal policies. Moreover, we provide recommendations to local governments on fuel station construction and vehicle subsidy policies, contributing practical solutions to both enterprise operations and regional development.}, archive = {J_ESWA}, author = {Zhang Yanhu and Yan Lijuan and Kong ShuMei and Miao Decheng}, doi = {10.1016/j.eswa.2025.129595}, journal = {Expert Systems with Applications}, month = {3}, pages = {129595}, shortjournal = {Expert Syst. Appl.}, title = {Research on multi-objective optimization of multi-endpoint VRP with time window for the distribution of seasonal products by multi-homing heterogeneous fleets}, volume = {298}, year = {2026}, }
• (2026). Revealing the correlation between economic indicators and gold prices for forecasting: Medium term forecast framework with data patch. ESWA, 298, 129594. (www)

  @article{ , abstract = {Predicting gold prices through the analysis of key economic indicators such as inflation rates, Government Bond Yields, and the U.S. Dollar Index, alongside historical Gold Prices, is crucial for enabling investors to better understand market dynamics and make vital decisions to maximize returns. However, previous studies have faced challenges in extracting hidden factors related to gold price prediction from diverse economic indicators, and the comprehensive exploration of gold price data is yet to be fully achieved. To address this, the present study introduces a mid to long-term gold price prediction model named DPformer. This model utilizes a patching strategy to investigate the relationships between different economic indicators and Gold Prices. It also employs a decomposition approach to discover the mid to long-term trend characteristics and yearly seasonal patterns of Gold Prices. The core of the model integrates a Transformer module, which is solely based on an Encoder structure, and enhances it with multiple attention mechanisms and convolutions. This enhancement allows the improved Transformer model to more effectively capture the long-term dependencies of Gold Prices. The empirical results demonstrate that DPformer consistently outperforms a suite of advanced models widely adopted in terms of mid to long-term forecasting accuracy, including LSTM, GRU, Transformer, DLinear, and PatchTST. Notably, for the 30-step gold price prediction task, DPformer achieves a 21.78 % reduction in Mean Squared Error compared to PatchTST. Moreover, by quantitatively analyzing how various economic indicators influence gold price forecasts, this study provides substantial support for investors in making informed decisions at critical moments.}, archive = {J_ESWA}, author = {Guanhao Bao and Yunbo Niu and Baisheng Cui and Wanying Ji}, doi = {10.1016/j.eswa.2025.129594}, journal = {Expert Systems with Applications}, month = {3}, pages = {129594}, shortjournal = {Expert Syst. Appl.}, title = {Revealing the correlation between economic indicators and gold prices for forecasting: Medium term forecast framework with data patch}, volume = {298}, year = {2026}, }
• (2026). Overview of machine learning in class imbalance scenarios: Trends, challenges, and approaches. ESWA, 298, 129592. (www)

  @article{ , abstract = {This study presents a systematic mapping of machine learning in class imbalance scenarios, offering a broad overview of key challenges, promising emerging techniques, and established methodologies across various application domains. The investigation stands out by employing a hybrid search and selection protocol that combines methodological rigor with technical innovation. The adopted strategy integrated manual searches in reference sources with automated processes based on machine learning, semantic embeddings, and graph-based ranking algorithms. To enhance selection quality, the Quasi-Golden Set (QGS) method was used to build a reference set from manually selected articles – a critical foundation for calibrating and evaluating automated search strings. This combination ensured broad coverage of the topic while improving sensitivity and precision in identifying relevant studies. The initial analysis reviewed 25,593 publications. After screening and applying eligibility criteria, 468 articles were included in the final dataset. The results indicate that 55 % of the studies address multiple domains, with a strong predominance of tabular data ( 84 % ). SMOTE and hybrid approaches were among the most common techniques, present in 61 % of the studies. In terms of evaluation metrics, ROC-AUC was the most frequently used, followed by F1-score and accuracy – the latter noted for limitations in highly imbalanced scenarios. Building on these findings, we derive an empirically grounded taxonomy that links problem context, solution algorithms, and scenario-appropriate evaluation metrics, and we provide a minimal selection guideline table to support applied use. While sampling-based methods remain prevalent, deep learning approaches such as convolutional neural networks and graph-based models are increasingly adopted. Additionally, federated, contrastive, and semi-supervised learning are emerging as relevant paradigms, particularly suited for privacy-aware or low-label environments. This study consolidates current knowledge, identifies methodological and application gaps, and highlights trends that are likely to shape future research. It contributes both a comprehensive synthesis of the field and strategic insights for advancing machine learning techniques in the presence of class imbalance.}, archive = {J_ESWA}, author = {Gilberto Sussumu Hida and André Câmara Alves Do Nascimento}, doi = {10.1016/j.eswa.2025.129592}, journal = {Expert Systems with Applications}, month = {3}, pages = {129592}, shortjournal = {Expert Syst. Appl.}, title = {Overview of machine learning in class imbalance scenarios: Trends, challenges, and approaches}, volume = {298}, year = {2026}, }
• (2026). Multi-objective multi-product process planning with reconfigurable machines: Exact and metaheuristic approaches. ESWA, 298, 129591. (www)

  @article{ , abstract = {Process planning in reconfigurable manufacturing systems usually considers a single product, this reduces the efficiency of the overall production plan when multiple products are combined. This paper tackles the Multi-Product Process Planning Problem (MPPP), optimizing both individual process plans and their sequencing. We propose a 0–1 LP model, the model is relaxed by fixing the product sequencing variables and implemented in a Normal-Boundary Intersection method (NBI-es), the method uses a function for iteratively updating β values. Three metaheuristics are also developed: NSGA-II, and two variants of MOEA/D, one enhanced by Opposition-based learning (OBL). Computational experiments show that the update function enhances the performance of NBI over simple Normalized-Weighted Sum (NWS) method. Additionally, NBI-es performs better in HV metric for small size instances if it is given enough CPU times, while MOEA/D significantly outperforms NSGA-II on larger instances on most convergence and spread based metrics. OBL further enhances solution diversity for MOEA/D, albeit with less convergence. A special case of the MPPP is investigated, involving identical products: the Multi-Unit Process Planning (MUPP). An integrated approach was compared with a sequential separated approach. Results indicate that the integrated approach outperforms the separated method for smaller problem instances. Moreover, the analysis of high-quality MUPP solutions revealed a tendency towards diverse process plan combinations rather than repetitive identical ones.}, archive = {J_ESWA}, author = {Abdelkader Mechaacha and Fayçal Belkaid and Nadjib Brahimi}, doi = {10.1016/j.eswa.2025.129591}, journal = {Expert Systems with Applications}, month = {3}, pages = {129591}, shortjournal = {Expert Syst. Appl.}, title = {Multi-objective multi-product process planning with reconfigurable machines: Exact and metaheuristic approaches}, volume = {298}, year = {2026}, }
• (2026). A slack-based two-stage improved particle swarm optimization algorithm for robust scheduling of a flexible job-shop with new and remanufacturing jobs. ESWA, 298, 129590. (www)

  @article{ , abstract = {Remanufacturing has attracted increasing attention for its environmental and economic benefits. Since it is difficult to achieve economies of scale when processing small amounts of remanufacturing jobs alone, these jobs are processed in the same job-shop for new jobs in some enterprises. The processing times of remanufacturing jobs are uncertain due to unpredictable status, leading to certain impacts on scheduling performance. Therefore, we address a flexible job-shop scheduling problem with new and remanufacturing jobs to minimize makespan. To solve this problem, a slack-based two-stage improved particle optimization algorithm is proposed. The first stage aims to yield a solution set with minimum makespan, while the second stage aims to search the best robust solution with maximum total slack from the set. Both stages are executed alternately to optimize makespan and total slack. Moreover, a position updating mechanism with genetic operators and a tabu search inspired local search strategy are implemented to improve algorithmic performance. Computational experiments are conducted using adapted benchmark problems and an industrial case to validate the proposed algorithm.}, archive = {J_ESWA}, author = {Jun Liu and Zhui Gui and An Li and Qiong Liu}, doi = {10.1016/j.eswa.2025.129590}, journal = {Expert Systems with Applications}, month = {3}, pages = {129590}, shortjournal = {Expert Syst. Appl.}, title = {A slack-based two-stage improved particle swarm optimization algorithm for robust scheduling of a flexible job-shop with new and remanufacturing jobs}, volume = {298}, year = {2026}, }
• (2026). Evaluating and predicting green economic efficiency in chinese cities: A three-stage network SBM and machine learning approach. ESWA, 298, 129589. (www)

  @article{ , abstract = {This paper evaluates and predicts green economic efficiency (GEE) across 248 Chinese cities from 2010 to 2021 using a three-stage network SBM model based on subsystems of economic production, social development, and environmental governance. To enhance accuracy in both assessment and forecasting, machine learning methods are incorporated, and the Dagum Gini coefficient is employed to analyze regional disparities. This study innovatively proposes a three-stage network SBM model to resolve the “black box” limitation of conventional DEA approaches, while a DEA-ML model is developed to achieve enhanced prediction accuracy. The results reveal that GEE in Chinese cities remains low, with the eastern region leading and the western region trailing. However, efficiency has improved since 2016, primarily driven by advancements in environmental governance. Regional disparities, largely attributed to interregional differences, are gradually decreasing. Among forecasting models, the backpropagation neural network (BPNN) delivers the highest accuracy, predicting sustained leadership in the east, strong growth in the northeast, and a reduction in national disparities. This study offers a comprehensive framework for evaluating and predicting GEE, providing valuable insights for sustainable development policies.}, archive = {J_ESWA}, author = {Zhishuo Zhang and Hu Liu and Yunpeng Gong and Huayong Niu}, doi = {10.1016/j.eswa.2025.129589}, journal = {Expert Systems with Applications}, month = {3}, pages = {129589}, shortjournal = {Expert Syst. Appl.}, title = {Evaluating and predicting green economic efficiency in chinese cities: A three-stage network SBM and machine learning approach}, volume = {298}, year = {2026}, }
• (2026). HAEA: A heterogeneous alternating evolutionary algorithm for numerical optimization. ESWA, 298, 129587. (www)

  @article{ , abstract = {A hybrid algorithm integrating a couple of individual evolutionary algorithms (sub-algorithms) is widely recognized as an effective approach to enhance both robustness and optimization performance. Nevertheless, such integration often destroys the structure of the sub-algorithm and makes it difficult to incorporate additional evolutionary algorithms. To address these limitations, this study introduces a novel framework, the Heterogeneous Alternating Evolutionary Algorithm (HAEA), designed to integrate multiple evolutionary algorithms while enabling the flexible addition, removal, and replacement of internal sub-algorithms. To facilitate the integration of a broad spectrum of sub-algorithms, this study draws inspiration from the particle swarm optimization algorithm to devise a suite of information indicators for the transmission of optimization information between sub-algorithms with disparate structures. Furthermore, HAEA is endowed with an adaptive mechanism that dynamically modifies the selection probabilities of its sub-algorithms based on their long-term and short-term performance throughout the evolutionary process. We conducted a comparative analysis of HAEA against all its sub-algorithms across three widely recognized function test sets: CEC2013, CEC2017, and CEC2022. Meanwhile, we applied the HAEA separately to basic metaheuristic algorithms and advanced evolutionary algorithms in recent years and conducted two comparative experiments. Both experimental results show that HAEA outperforms all sub-algorithms in terms of robustness and optimization performance. Its distinctive flexibility allows for the incorporation of additional superior evolutionary algorithms in the future, thereby enhancing its overall performance.}, archive = {J_ESWA}, author = {Taiyong Li and Tianhao Yi and Zhenda Hu and Wu Deng and Donglin Zhu and Zhilong Xie and Jiang Wu}, doi = {10.1016/j.eswa.2025.129587}, journal = {Expert Systems with Applications}, month = {3}, pages = {129587}, shortjournal = {Expert Syst. Appl.}, title = {HAEA: A heterogeneous alternating evolutionary algorithm for numerical optimization}, volume = {298}, year = {2026}, }
• (2026). Deep analysis on MLSY for fractional-order higher-dimension-valued neural networks under the action of free quadratic coefficients. ESWA, 298, 129586. (www)

  @article{ , abstract = {In this article, free quadratic coefficients are proposed in order to deeply study the flexible criteria of synchronization problem for two kinds of fractional-order higher-dimension-valued neural networks (FOHDVNN) with usual neurons and threshold ones, respectively. First, the uniform system is constructed for two kinds of FOHDVNN which contains both fractional-order octonion-valued neural networks (FOOVNN) and fractional-order quaternion-valued neural networks (FOQVNN). Based on higher-dimension algebra multiplication rules, the studied FOHDVNN are directly decomposed into the eight or four subsystems in real-valued field. Subsequently, free quadratic coefficients are taken into the establishment of two types of Lyapunov-Krasovskii functional (LKF) which is newer and more general. Then, mainly based on the very recent lemmas and Lyapunov theories, the flexible criteria are generally acquired for the global Mittag-Leffler synchronization (MLSY) problem of FOHDVNN. The final criteria have the advantage in being easily calculated and widely used. It is worth noting that the optimal solutions of these criteria can be obtained through the genetic algorithm and the synchronization performance can be improved by optimizing the quadratic coefficients. Finally, three simulation examples are presented to express the availability and progress of the derived results.}, archive = {J_ESWA}, author = {Jianying Xiao and Yongtao Li}, doi = {10.1016/j.eswa.2025.129586}, journal = {Expert Systems with Applications}, month = {3}, pages = {129586}, shortjournal = {Expert Syst. Appl.}, title = {Deep analysis on MLSY for fractional-order higher-dimension-valued neural networks under the action of free quadratic coefficients}, volume = {298}, year = {2026}, }
• (2026). Complex-order darwinian particle swarm optimization. ESWA, 298, 129584. (www)

  @article{ , abstract = {The Particle Swarm Optimization (PSO) algorithm has been one of the most effective methods for solving various complex optimization problems. However, non-adaptive versions of the PSO do not use historical information for performance enhancement and suffer from performance degradation problems. This paper presents a Complex-Order Darwinian PSO (CoDPSO) algorithm, which effectively enhances the performance of the PSO. A complex-order derivative mechanism is introduced into the velocity update rule to improve local exploitation using historical velocity information. Additionally, a Degradation Elimination (DE) strategy is designed to mitigate performance drop during the optimization process. Sensitivity analysis is conducted to evaluate the impact of control parameters on the algorithm’s behavior, demonstrating its robustness across a wide range of settings. Comparative experiments on CEC 2022 benchmark functions show that the CoDPSO outperforms other PSO variants in terms of accuracy, stability, and convergence. Wilcoxon statistical tests further confirm the significance of these improvements. The experimental results indicate the feasibility and efficiency of the CoDPSO.}, archive = {J_ESWA}, author = {Xiaobo Wu and Liping Chen and Huafeng Li and António M. Lopes and Chuang Liu and Yangquan Chen and Yi Chai}, doi = {10.1016/j.eswa.2025.129584}, journal = {Expert Systems with Applications}, month = {3}, pages = {129584}, shortjournal = {Expert Syst. Appl.}, title = {Complex-order darwinian particle swarm optimization}, volume = {298}, year = {2026}, }
• (2026). A multi-objective framework for predicting public opinion trends on infectious diseases using NSGA-II and interval predictions. ESWA, 298, 129583. (www)

  @article{ , abstract = {Predicting public opinion trends during major infectious disease outbreaks is critical for guiding effective public health responses. However, predicting public opinion remains challenging because it is influenced by socio-economic, psychological, and media factors. This paper presents a novel framework for predicting public opinion trends related to significant infectious diseases, with a focus on COVID-19 as a case study. The proposed framework identifies the key factors influencing public opinion development and enables both point and interval predictions. The framework uses information ecology theory and applies the NSGA-II algorithm to select the features that best drive public opinion trends. By incorporating this framework, accurate point forecasts are produced alongside prediction intervals, effectively quantifying the uncertainty inherent in public opinion dynamics. This approach minimizes the quality-driven loss function to generate precise prediction intervals, providing decision-makers with critical insights into public opinion fluctuations during epidemics. The results offer valuable, real-time public sentiment warnings, supporting timely and effective interventions in epidemic prevention and control efforts.}, archive = {J_ESWA}, author = {Futian Weng and Meng Su and Petr Hajek and Mohammad Zoynul Abedin}, doi = {10.1016/j.eswa.2025.129583}, journal = {Expert Systems with Applications}, month = {3}, pages = {129583}, shortjournal = {Expert Syst. Appl.}, title = {A multi-objective framework for predicting public opinion trends on infectious diseases using NSGA-II and interval predictions}, volume = {298}, year = {2026}, }
• (2026). A deep reinforcement learning-guided multimodal multi-objective evolutionary algorithm with a serial-parallel mechanism. ESWA, 298, 129581. (www)

  @article{ , abstract = {The core challenge for multimodal multi-objective problem (MMOP) resolution lies in maintaining synergistic interactions between convergence and diversity. However, the existing algorithms usually consider convergence-first, which neglect to consider both diversity and convergence into account during the evolutionary process. Likewise, the optimization methods tend to gravitate toward locally optimal regions rapidly, leading to lose diversity for the local PS. This paper proposes a Deep Reinforcement Learning-guided multimodal multi-objective evolutionary algorithm with a serial-parallel mechanism (DRLMMEA) to investigate the impact of different operator selection on the performance of MMEAs, which greatly helps to balance the convergence and diversity. DRLMMEA utilizes Q-Network to select the operator with the highest reward to enhance the population’s search ability. An improved sorting method (ISM) based on neighborhood dominance updates the population by sorting individuals according to their convergence quality, thereby enhancing convergence performance in the objective space. Moreover, this study proposes a series-parallel mechanism, a series structure enhances the diversity in the decision space, while the parallel structure reduces the computational burden of the algorithm evidently. The proposed Deep Reinforcement Learning-assisted operator selection mechanism, which enables effective balance between diversity and convergence, and an improved crowding distance approach that enhances convergence performance. DRLMMEA undergoes comprehensive testing against 6 contemporary approaches using MMF and IDMP benchmark problems, achieving supremacy in 4 principal performance metrics according to experimental findings. The multimodal gearbox parameter optimization is addressed using the proposed DRLMMEA, which demonstrates superior performance against 6 algorithms in comparative evaluations. It has demonstrated a significant role in solving the MMOPs with the imbalance between convergence and diversity.}, archive = {J_ESWA}, author = {Ying Huang and Xiaojian Cao and Benben Zhou and Wei Li and Shuling Yang and S.M. Shafi and Zhou Yang}, doi = {10.1016/j.eswa.2025.129581}, journal = {Expert Systems with Applications}, month = {3}, pages = {129581}, shortjournal = {Expert Syst. Appl.}, title = {A deep reinforcement learning-guided multimodal multi-objective evolutionary algorithm with a serial-parallel mechanism}, volume = {298}, year = {2026}, }
• (2026). Evolutionary multitasking optimization based on cross-task association mapping strategy. ESWA, 298, 129580. (www)

  @article{ , abstract = {In multitasking optimization, knowledge transfer between tasks through subspace generation has been widely employed to enhance the convergence performance of algorithms. However, this approach fails to account for the inter-task knowledge mapping relationships. Therefore, cross-task knowledge transfer during the optimization process remains inherently blind, potentially leading to mismatched subspace information and consequently degrading the algorithm’s performance. To address this issue, this paper proposes a multitask evolutionary algorithm based on an association mapping strategy and an adaptive population reuse mechanism, namely PA-MTEA. Specifically, to fully represent the correlations between multitask domains and enhance the adaptability of transfer solutions in target tasks, this paper introduces a subspace projection strategy based on partial least squares, which achieves the correlation mapping between the source and target tasks during the dimensionality reduction of the search space. Additionally, to further enhance knowledge transfer across tasks, an alignment matrix is obtained by adjusting the subspace Bregman divergence after deriving the respective subspaces, minimizing variability between task domains. Finally, to balance the global exploration of algorithms with local exploitation, an adaptive population reuse mechanism based on the residual structure is designed. This mechanism reuses historically successful individuals to guide the evolutionary direction of the population, thus improving the algorithm’s convergence performance. Experimental results on various benchmark suites and real-world cases demonstrate that PA-MTEA exhibits significantly superior performance compared to six other advanced multitask optimization algorithms.}, archive = {J_ESWA}, author = {Tao Yin and Lizhong Yao and Xin Zong and Pengjie Qin and Haoming Dong}, doi = {10.1016/j.eswa.2025.129580}, journal = {Expert Systems with Applications}, month = {3}, pages = {129580}, shortjournal = {Expert Syst. Appl.}, title = {Evolutionary multitasking optimization based on cross-task association mapping strategy}, volume = {298}, year = {2026}, }
• (2026). UniTask+: Exploring and unifying strong and weak task-aware consistency for semi-supervised blastocyst image segmentation. ESWA, 298, 129578. (www)

  @article{ , abstract = {The segmentation of different tissues within blastocysts is essential for embryologists to objectively observe and evaluate embryos, thereby contributing to a higher success rate of in vitro fertilization treatment. Inspired by the primary observation of skeletal patterns and boundary information by clinical doctors, we present an interesting task-aware view for blastocyst segmentation with semi-supervised learning, focusing on task-invariant and task-specific dependencies of segmentation. Firstly, we explore one strong-correlation task with bidirectional transformation between its outputs and the segmentation results, and another weak-correlation task with monodirectional transformation from segmentation maps. The correlation among different tasks inspires us to propose Task-Aware Smoothness (TAS) Assumption , thereby deducing different types of task-aware consistency. Then, a new Unified Task-aware Consistency Interaction (UniTask+) framework is developed to unify and fully take advantage of these strong, weak, and strong-to-weak task-aware consistency. It is comprised of a medical segmentation (MS) branch to implement segmentation and two extra branches performing strong/weak-correlation tasks based on the same backbone. Concretely, a level-set (LS) branch promotes the strong consistency while a point-set (PS) branch stimulates the weak consistency with underlying task perturbations. Numerous experiments have been conducted on the inner cell mass (ICM), blastocyst, proving the effectiveness of our tactics. Furthermore, we have also conducted experiments with datasets from the left atrium (LA), which shares similar structural features with embryos, to validate the robustness of the model. Our methods have shown prominent improvements over up-to-date SSL methods, which advocates our precedent hypothesis.}, archive = {J_ESWA}, author = {Hua Wang and Linwei Qiu and Jingfei Hu and Jicong Zhang}, doi = {10.1016/j.eswa.2025.129578}, journal = {Expert Systems with Applications}, month = {3}, pages = {129578}, shortjournal = {Expert Syst. Appl.}, title = {UniTask+: Exploring and unifying strong and weak task-aware consistency for semi-supervised blastocyst image segmentation}, volume = {298}, year = {2026}, }
• (2026). Deep asymmetric semantic hashing with probability shifting for multi-label image retrieval. ESWA, 298, 129577. (www)

  @article{ , abstract = {Deep hashing has been widely used in large-scale multimedia retrieval due to its advantages in terms of low storage cost and computational efficiency. Deep hashing algorithms can jointly learn semantic features and hash functions, encoding the original data into compact binary codes with significant discriminative power. However, in multi-label scenarios, especially when the number of samples is extremely large, a high negative-positive imbalance may occur, particularly when the proportion of negative samples is too high, which can lead to bias in the semantic relationships between the learned images. To solve this problem, symmetric losses such as focal loss were proposed, which treat positive and negative samples equally, but the retrieval results are suboptimal. This may be because the equal-weighted processing strategy causes the model to over-focus on hard negative samples and ignore the learning of positive sample features. Besides, mislabeled negative samples, especially those with a probability close to 1, can lead the model to learn incorrect features, harming its discrimination ability, reducing accuracy and recall, and causing overfitting and poor generalization. Accordingly, this paper proposes a novel hashing model, Deep Asymmetric Semantic Hashing with Probability Shifting framework (DASH-PS), for discriminative binary code learning. Specifically, by combining asymmetric focusing strategy and probability shifting strategy, asymmetric semantic loss is designed to solve negative-positive imbalance and ground-truth mislabeling. To keep the contribution of positive samples while focusing on hard negative samples, asymmetric focusing strategy is proposed to decouple negative and positive samples and assign different attenuation factors. By offsetting the probability of negative samples, probability shifting strategy completely discards easy negative samples and very hard negative samples suspected of being mislabeled. Additionally, an adaptive asymmetric learning mechanism is proposed to reduce the fixed difference in average probabilities between positive and negative samples, thereby simplifying hyperparameter selection and improving retrieval efficiency. Extensive experimental results on multiple benchmark datasets validate that our DASH-PS outperforms various state-of-the-art hashing methods. The code for the implementation of our DASH-PS framework is available at https://github.com/QinLab-WFU/DASH-PS.}, archive = {J_ESWA}, author = {Yongyue Fu and Qibing Qin and Jinkui Hou and Congcong Zhu and Lei Huang and Wenfeng Zhang}, doi = {10.1016/j.eswa.2025.129577}, journal = {Expert Systems with Applications}, month = {3}, pages = {129577}, shortjournal = {Expert Syst. Appl.}, title = {Deep asymmetric semantic hashing with probability shifting for multi-label image retrieval}, volume = {298}, year = {2026}, }
• (2026). Physics-informed tensor autoencoder with memory for video anomaly detection. ESWA, 298, 129576. (www)

  @article{ , abstract = {Video data can be naturally represented as tensors. Despite great progress in anomaly detection with memory-augmented autoencoders, the memory module therein can only handle vectors and inevitably breaks tensor structures, thus leading to performance degradation. Moreover, after the mapping of the encoder, some abnormal features may directly fall into the normal convex polytope, as autoencoders only use the output error to guide the construction of latent variables without imposing any constraint. The memory module can not handle these abnormal features, so that the abnormal observations may not be identified. To solve these problems, we propose a Physics-Informed Tensor AutoEncoder (PITAE) framework, which incorporates both neural networks and physical laws, i.e., tensor operation rules. Specifically, we design a tensor decomposition network followed by an explicit tensor operation to decompose the latent variable into low-rank and sparse components, and only the low-rank component is inputted to the decoder. In this way, we reserve the tensor structure and meanwhile impose a low-rank constraint on the latent variable, thereby compressing the features of normal samples into a ”smaller” region where anomalies are less likely to fall into. Consequently, the non-low-rank anomalies can be identified. But the low-rank anomalies may still not be identified. To further solve this problem, we design a tensor Memory module, and the overall model is named as PITAEM. Finally, based on the proposed framework, we design a novel composite anomaly score to identify anomalies of various kinds. Experiments on various video datasets demonstrate the effectiveness of the proposed method, especially in the small data regime.}, archive = {J_ESWA}, author = {Jianan Liu and Chunguang Li}, doi = {10.1016/j.eswa.2025.129576}, journal = {Expert Systems with Applications}, month = {3}, pages = {129576}, shortjournal = {Expert Syst. Appl.}, title = {Physics-informed tensor autoencoder with memory for video anomaly detection}, volume = {298}, year = {2026}, }
• (2026). A robust medical image encryption technique using inverse cosine chaotic map. ESWA, 298, 129574. (www)

  @article{ , abstract = {With the rapid proliferation of digital imaging technologies, the need for robust and lightweight image encryption techniques has become increasingly critical, particularly for medical, military, and personal data applications. In this paper, we propose a novel image encryption scheme based on a one-dimensional inverse cosine chaotic map (1D-ICC), which introduces a highly sensitive and structurally complex nonlinear dynamical system. The proposed method integrates a dynamic Josephus-based intra-block scrambling mechanism, a global zigzag permutation strategy, and an adaptive diffusion process guided by chaotic sequences, thereby enhancing the confusion and diffusion characteristics of the cipher. Unlike conventional approaches, our scheme dynamically derives the encryption key from the SHA-512 hash of the original image, ensuring both sensitivity to plaintext changes and resistance to known-plaintext and chosen-plaintext attacks. The use of the 1D-ICC map, featuring a tunable control parameter r 5 enables rich chaotic behavior even in one dimension, reducing computational complexity without sacrificing security. Comprehensive experiments validate the robustness and efficiency of the encryption scheme, with performance metrics including correlation coefficients below 0.003, information entropy of 7.9993, a Number of Pixels Change Rate (NPCR) of 99.61 %, and a Unified Averaged Changed Intensity (UACI) of 33.42 %. These results demonstrate that our method surpasses several existing techniques in both security strength and computational performance, underscoring the potential of the 1D-ICC map for practical image encryption applications.}, archive = {J_ESWA}, author = {Jackson J and Perumal R}, doi = {10.1016/j.eswa.2025.129574}, journal = {Expert Systems with Applications}, month = {3}, pages = {129574}, shortjournal = {Expert Syst. Appl.}, title = {A robust medical image encryption technique using inverse cosine chaotic map}, volume = {298}, year = {2026}, }
• (2026). Enhancing text classification with neural label embedding and weakly-supervised learning. ESWA, 298, 129569. (www)

  @article{ , abstract = {Recent years have witnessed the widespread adoption of deep-learning-based models in a range of linguistic tasks including the fundamental text classification. These deep neural networks, however, often face challenges due to the limited availability of large-scale training data with high-quality label annotations. Furthermore, while supervised learning has proven to be superior in training sentence representations for downstream tasks like text classification, this aspect has received relatively little attention. In this study, a novel model named L abel Em bedding joint with We akly-supervised C lassification ( LemWec ) is proposed, which aims to establish a unified framework by combining supervised sentence embedding with multiclass classification. For supervised sentence embeddings, the model incorporates seed information such as label names and designs an encoder network with a new pooling layer. Additionally, the model adopts a pseudo-labeling approach to leverage a substantial amount of unlabeled samples. This approach specifically addresses the drawback of generating pseudo-labels with the highest confidence and introduces a noise adaptation method to mitigate this issue. The results of extensive experiments conducted on four real-world datasets demonstrate that the proposed LemWec model can significantly enhance the performance of text classification when compared to a comprehensive set of baselines.}, archive = {J_ESWA}, author = {Xiao Jing and Zhe Li and Zhiang Wu and Dejun Mu}, doi = {10.1016/j.eswa.2025.129569}, journal = {Expert Systems with Applications}, month = {3}, pages = {129569}, shortjournal = {Expert Syst. Appl.}, title = {Enhancing text classification with neural label embedding and weakly-supervised learning}, volume = {298}, year = {2026}, }
• (2026). Self-supervised ZINB-based kernel representation learning of single-cell RNA sequencing data. ESWA, 298, 129568. (www)

  @article{ , abstract = {Cell clustering plays a vital role in single-cell RNA sequencing (scRNA-seq) data analysis. Although many deep cell clustering methods have been proposed to cluster the scRNA-seq data, they overlook the structural partitioning objectives during the representation learning process, leading to challenges with non-linearly separable structures. In this paper, we present a novel end-to-end deep kernel cell clustering model for scRNA-seq data based on self-supervised ZINB-based kernel representation learning, named scDKC, which simultaneously learns cell kernel representations and identifies cell clusters. Specifically, a kernel-aid hybrid representation learning encoder is developed to effectively learn the separable kernel representation of cells, consisting of cells’ expression characteristics and cell-cell topological interactions. To guide the direction of kernel representation learning, a ZINB-based kernel representation learning decoder is designed by capturing the global probabilistic structure, the representation and the cell graph structure of the scRNA-seq data. By leveraging the clustering self-supervised strategy, representation self-supervised strategy, ZINB-based distribution self-supervised strategy, and kernel self-supervised strategy, scDKC optimizes cell cluster label assignment and learns cell kernel representations through a joint mutual self-supervised mechanism. Extensive experiments on 15 real scRNA-seq datasets, comparing scDKC with 10 competing methods, highlight its competitive advantages.}, archive = {J_ESWA}, author = {Lina Ren and Maoxuan Yao and Ruizhang Huang and Yongbin Qin}, doi = {10.1016/j.eswa.2025.129568}, journal = {Expert Systems with Applications}, month = {3}, pages = {129568}, shortjournal = {Expert Syst. Appl.}, title = {Self-supervised ZINB-based kernel representation learning of single-cell RNA sequencing data}, volume = {298}, year = {2026}, }
• (2026). A dual uncertainty-aware fusion framework for face expression recognition in the wild. ESWA, 298, 129567. (www)

  @article{ , abstract = {Facial Expression Recognition(FER) is a key task in the broader landscape of affective computing and human-computer interaction, enabling machines to interpret human emotions. To better learn discriminative features under complex facial variations, recent FER research has increasingly adopted multi-branch fusion architectures that aim to capture complementary features from diverse perspectives. However, existing multi-branch fusion strategies, including static weighting, simple concatenation, or uncertainty-aware modeling, lack the capacity to comprehensively capture and reconcile the reliability variations across both individual instances and structural branches. To overcome these limitations, we propose a novel multi-branch fusion strategy, named Dual Uncertainty-Aware Fusion Framework(DUAFF), which improves the discriminability of integrated features by simultaneously modeling instance-wise uncertainty and inter-branch correlations. Specifically, the proposed method comprises two complementary modules: Instance-Discrepant Uncertainty-Aware Fusion Module (ID-UAFM) and Branch-Discrepant Uncertainty-Aware Fusion Module (BD-UAFM). ID-UAFM is introduced to perform channel-wise entropy analysis between semantically distinct samples to estimate instance-level uncertainty, enabling selective channel-wise fusion that emphasizes reliable representations while suppressing uncertain responses. BD-UAFM is further proposed to capture structural uncertainty by evaluating the relative reliability of features across multiple branches and adaptively weighting their contributions based on inter-branch discrepancies. Experimental results demonstrate that the proposed DUAFF consistently outperforms POSTER across three benchmark datasets, achieving accuracy improvements of 0.23 % on RAF-DB, 0.69 % on FER2013, and 0.29 % on AffectNet (7-class), thereby confirming its effectiveness in enhancing the reliability and discriminability of facial representations.}, archive = {J_ESWA}, author = {Wenfeng Jiang and Ziyi Zhao and Lin Wang and Fang Liu and Chunmei Qing and Xiaofen Xing and Xiangmin Xu and Weiquan Fan and Zhanpeng Jin}, doi = {10.1016/j.eswa.2025.129567}, journal = {Expert Systems with Applications}, month = {3}, pages = {129567}, shortjournal = {Expert Syst. Appl.}, title = {A dual uncertainty-aware fusion framework for face expression recognition in the wild}, volume = {298}, year = {2026}, }
• (2026). Forecasting tourism stock index dynamics: A multiscale deep learning framework integrating emerging media data. ESWA, 298, 129566. (www)

  @article{ , abstract = {This study introduces a novel multiscale and multivariable deep learning framework for tourism stock index forecasting. To address the research gap concerning emerging media’s impact on the tourism sector, our study innovatively integrate multi-source data, including Douyin (China’s prominent short video platform), into our predictive model. Our methodology employs a multiscale decomposition strategy to streamline feature extraction complexity, coupled with an enhanced temporal convolutional network model incorporating soft-thresholding denoising to mitigate noise interference. Furthermore, we implement an adaptive differentiated prediction strategy to optimize model flexibility. Empirical analysis utilizing the CSI Tourism Stock Index demonstrates that our proposed model outperforms benchmark models in both predictive accuracy and stability, thereby validating its efficacy in tourism stock index forecasting.}, archive = {J_ESWA}, author = {Feng Shen and Shuai Huang and Wanqing Zhao and Dao Lan}, doi = {10.1016/j.eswa.2025.129566}, journal = {Expert Systems with Applications}, month = {3}, pages = {129566}, shortjournal = {Expert Syst. Appl.}, title = {Forecasting tourism stock index dynamics: A multiscale deep learning framework integrating emerging media data}, volume = {298}, year = {2026}, }
• (2026). Enhancing adversarial transferability through frequency-domain boundary samples tuning. ESWA, 298, 129565. (www)

  @article{ , abstract = {Transfer-based attacks evaluate the robustness of deep learning models and advance adversarial research to improve the security and reliability of deep learning and its applications. Previous efforts have improved the transferability through advanced gradients, augmented models, or augmented data. In this paper, we understand and enhance the transferability from a new perspective. Delving into intermediate features, we empirically find a difference between the distances of adversarial and original samples from cluster centers of the original classes. The adversarial samples are simultaneously far from both the original samples and the cluster centers, close to generalized decision boundaries. Based on this observation, we propose a novel spectrum tuning attack. Boundary samples are utilized to guide the generation of adversarial samples that are far away from the cluster centers. Specifically, randomized boundary samples are generated by frequency-domain transformations. With the gradients of the diverse boundary samples, the adversarial perturbation moves the examples away from cluster centers, thus approaching generalized decision boundaries. In the optimization process, conjugate directions are employed to avoid oscillations and stabilize the update direction. Given the strong Wolfe parameters, the analysis of the descent direction and current gradient further ensures the convergence speed and stability of the optimization. In addition, Gaussian preprocessing is introduced to smooth the update direction, further stabilize the direction and enhance the transferability. The proposed method is flexible enough to be combined with existing methods to further improve the transferability. Experiments conducted on the ImageNet-compatible dataset validate the effectiveness of the proposed method, e.g., 92.6 % success rate against nine defense methods.}, archive = {J_ESWA}, author = {Shuyan Cheng and Peng Li and Keji Han and Yangjun Xiong and He Xu and Ruchuan Wang}, doi = {10.1016/j.eswa.2025.129565}, journal = {Expert Systems with Applications}, month = {3}, pages = {129565}, shortjournal = {Expert Syst. Appl.}, title = {Enhancing adversarial transferability through frequency-domain boundary samples tuning}, volume = {298}, year = {2026}, }
• (2026). When graph anomaly breaks the coherence: A multi-evidence approach with language models. ESWA, 298, 129557. (www)

  @article{ , abstract = {Graph anomaly detection is critical in domains such as healthcare and economics, where identifying deviations can prevent substantial losses. However, current detection methods, primarily reliant on Graph Neural Networks (GNNs), suffer from a critical limitation: they make judgment on a single piece of evidence – the classification of learned node representations. This “single-verdict” approach is inherently susceptible to misjudgments arising from noisy or biased representations. To address this limitation, we introduce Multi-AD, a novel Multi-evidence-based graph Anomaly Detection framework that leverages the power of Language Models (LMs) to enable more robust and reliable anomaly detection. We provide a paradigm shift by constructing multiple evidence sequences for each target node, and employing LMs to assess the coherence of these sequences. By aggregating coherence scores across multiple sequences, Multi-AD leverages converging evidence to make more informed decisions about anomaly status as the presence of anomalous nodes disrupts coherence. Furthermore, we introduce a coherence-aware edge representation method to enhance the discriminative power of the constructed sequences and a multi-round adaptive integration strategy to handle challenging scenarios where normal nodes might be surrounded by anomalies. Extensive experiments demonstrate that Multi-AD consistently outperforms state-of-the-art methods.}, archive = {J_ESWA}, author = {Xuan Cheng and Jiahui Lu and Chunjing Xiao and Meiyi Yang and Meihui Zhong and Fan Zhou}, doi = {10.1016/j.eswa.2025.129557}, journal = {Expert Systems with Applications}, month = {3}, pages = {129557}, shortjournal = {Expert Syst. Appl.}, title = {When graph anomaly breaks the coherence: A multi-evidence approach with language models}, volume = {298}, year = {2026}, }
• (2026). Multimodal joint subspace model for parkinson’s disease diagnosis. ESWA, 298, 129556. (www)

  @article{ , abstract = {Parkinson’s disease (PD) is an irreversible neurodegenerative disorder that significantly impacts patients’ lives. Accurate early diagnosis prediction is crucial for providing timely treatment to delay disease progression. However, current diagnostic methods predominantly rely on the experience and judgment of clinicians, introducing subjectivity and a lack of standardized, quantitative measures. Sparse subspace learning, as a machine learning technique, can extract critical information from multimodal data while addressing issues such as noise, high-dimensional complexity, and class imbalance. Our study utilizes longitudinal, multimodal neuroimaging data collected at multiple time points to develop a diagnostic model for PD. The approach involves extracting latent local features and leveraging deep learning techniques to generate a comprehensive global feature subset. Adaptive sparse selection is employed to reduce feature redundancy. Finally, support vector machine is used for classification and regression tasks, specifically for PD diagnosis and disease progression score prediction. Extensive experiments were conducted on the PPMI dataset, achieving an accuracy of 90.78 % for Scan Without Evidence of Dopaminergic Deficit (SWEDD) vs. Normal Control (NC) classification, 83.79 % for PD vs. NC, and 91.50 % for PD vs. SWEDD. The results demonstrate that the proposed method improves PD classification and prediction performance, showing promise for early diagnostic applications.}, archive = {J_ESWA}, author = {Haojie Song and Haijun Lei and Yukang Lei and Zhongwei Huang and Jiaqiang Li and Tianfu Wang and Peng Yang and Baiying Lei}, doi = {10.1016/j.eswa.2025.129556}, journal = {Expert Systems with Applications}, month = {3}, pages = {129556}, shortjournal = {Expert Syst. Appl.}, title = {Multimodal joint subspace model for parkinson’s disease diagnosis}, volume = {298}, year = {2026}, }
• (2026). A spectral relevance analysis approach to pattern recognition of financial time series. ESWA, 298, 129555. (www)

  @article{ , abstract = {Understanding patterns in financial time series is crucial for improving prediction accuracy in algorithmic trading and risk management. This paper presents a novel AI-based computer vision approach for classifying financial time series. Historical price sequences are transformed into Gramian Angular Difference Field (GADF) images and fed into a convolutional neural network (CNN) for pattern recognition. To interpret the CNN’s decision-making process, we apply Spectral Relevance Analysis (SpRAy), enabling the identification of distinct clusters based on relevance maps. Clustering the images according to their relevance profiles reveals groups with significantly higher predictive performance compared to the full dataset. The corresponding relevance patterns highlight favorable price movement structures and are identified via the associated clusters.}, archive = {J_ESWA}, author = {Christine Distler and Yarema Okhrin and Jonathan Pfahler}, doi = {10.1016/j.eswa.2025.129555}, journal = {Expert Systems with Applications}, month = {3}, pages = {129555}, shortjournal = {Expert Syst. Appl.}, title = {A spectral relevance analysis approach to pattern recognition of financial time series}, volume = {298}, year = {2026}, }
• (2026). A novel object detection system for computer night vision images using residual 3D transformer-based YoloV8 with adaptive GRU in edge and cloud sector. ESWA, 298, 129554. (www)

  @article{ , abstract = {Object identification is one of the computer vision-based methods used in locating and labelling objects in images. Object detection has been greatly advanced as it is now applicable for detecting night vision images with great accuracy. Most accurate object detection at night can be useful in many applications like nighttime driving, regulating harsh traffic in harsh weather conditions, and surveillance. Object detection in normal conditions can be smoother, but low illumination and harsh weather can lead to low versatility. Images captured at night can reflect a lot of noise with low visual features. Due to its challenging nature, a highly effective object detection model is a challenge for high-level applications. Traditional models still face issues and challenges related to uneven light conditions, brightness variations, different light sources, and noisy backgrounds that need to be addressed. Thus, it is necessary to develop an object detection model for dealing with images with low illumination and varying light conditions. Hence, in this work, an effective object detection framework is implemented for night vision images. At first, from the standard datasets, the significant night vision images are fetched and fed into the proposed model as a Residual 3D Transformer-based YoloV8 with an Adaptive Gated Recurrent Unit (R3DT-YAGRU) for detecting the objects. This includes combining spatial–temporal modelling capabilities into YOLOv8, particularly using 3D transformers for improved feature extraction and an adaptive GRU to manage temporal dependencies at night. Here, the Modified Random Variable-based Dollmaker Optimization Algorithm (MRV-DOA), which is a metaheuristic algorithm motivated by the doll-making procedure. Also, it helps in balancing the exploration phase and exploitation phase to discover the best solutions and is used for tuning the parameters of the R3DT-YAGRU model. At last, the experimental validation is carried out for the recommended object detection process by comparing with other models to establish the supremacy of the suggested work. From the study, the suggested framework achieves an accuracy of 96%, leading to enhanced decision making and better accuracy than other conventional models. The work has prepared its implementation accessible at https://github.com/charlesvprabhu56/Object-Detection .}, archive = {J_ESWA}, author = {V.Charles Prabu and Queen Mary Vidya. M and V. Sathiyamoorthi and P. Durgadevi and M. Gowthami}, doi = {10.1016/j.eswa.2025.129554}, journal = {Expert Systems with Applications}, month = {3}, pages = {129554}, shortjournal = {Expert Syst. Appl.}, title = {A novel object detection system for computer night vision images using residual 3D transformer-based YoloV8 with adaptive GRU in edge and cloud sector}, volume = {298}, year = {2026}, }
• (2026). Learning diversified features for pulmonary hypertension detection using chest X-ray. ESWA, 298, 129553. (www)

  @article{ , abstract = {Compared to traditional Computed Tomography (CT) scans and floatation catheters, chest X-ray offers an efficient, safe and timely examination paradigm, with broader range of scenarios (including intensive care units), for the detection of Pulmonary Arterial Hypertension (PAH). However, it is difficult to learn the variable radiological features of PAH from X-rays due to its low resolution and low contrast. To address the above issues, we propose a diversified features learning framework to fully explore the PAH-related representation from chest X-ray. We first employ a Chest Feature Enhancement Attention (CFEA) module to enhance the initial feature representation. Then, we employ the Deep Temporal Anti-Interference Metric Learning (TAIML) module to fully explore the PAH-related features. We incorporate the information on the temporal evolution of patients’ conditions. Specifically, a patient x , after undergoing treatment, may exhibit two possible states: x + (ill) and x − (cured). Therefore, we can define the distance d ( x , x + ) as the intra-class structural distance, and the distance d ( x , x − ) as the inter-class safe distance. Unlike existing metric learning, we adopt a new strategy: we push positive samples towards negative samples, but ensure distance between them is no less than d ( x , x − ) , thereby enhancing intra-class diversity while maintaining discriminability. Meanwhile, we ensure that the distance between positive samples is greater than d ( x , x + ) , thereby preserving the intra-class structure. Through these two steps, we can learn a diversified but discriminative representation of PAH. Comprehensive experiments showed the our model achieved an impressive accuracy of 86.27 % and an AUC of 0.857 in identifying PAH patients. The code is available at https://github.com/zgfdmn/PAH .}, archive = {J_ESWA}, author = {Chengjin Yu and Huanghui Wang and Yuanting Yan and Zhuyang Chu and Dongsheng Ruan}, doi = {10.1016/j.eswa.2025.129553}, journal = {Expert Systems with Applications}, month = {3}, pages = {129553}, shortjournal = {Expert Syst. Appl.}, title = {Learning diversified features for pulmonary hypertension detection using chest X-ray}, volume = {298}, year = {2026}, }
• (2026). TriHAtt-BERT-DBiLSTM: Deep learning model for medical emergency prediction from the unstructured data. ESWA, 298, 129552. (www)

  @article{ , abstract = {In recent years, the emergency triage has faced several challenges, including insufficient manual triage with physicians, limited medical resources contributing to incorrect triage, overcrowding in the emergency department (ED), and extended patient waiting time. Hence, the Medical Emergency prediction remains the major research area that identifies emergencies about specific diseases using the Medical Transcriptions (MT) provided by physicians. However, the existing methods face the challenges of handling the ambiguity of words, unstructured data, and increased computation complexity. Consequently, this research proposes the Tri-Head Attention-based Bidirectional Encoder Representations from Transformers enabled Distributed Bidirectional Long-Short Term Memory (TriHAtt-BERT-DBiLSTM) for predicting medical emergencies. Specifically, the proposed approach integrates the Tri-Head Attention mechanisms into BERT, which is further hybridized with the DBiLSTM model that offers the synergic strength of providing the dense feature representations to capture the complex dependencies, and enhancement of model ability with the structured parameters to facilitate the medical emergency prediction. Besides, the utilization of BERT in the proposed approach assists in capturing more complex language representations and further executes a better embedding representation of words. The TriHAtt-BERT-DBiLSTM model surpasses other state-of-the-art techniques and achieves 96.40% of accuracy, 96.12% of F1-score, 96.09% of precision, and 96.15% of recall for medical emergency prediction.}, archive = {J_ESWA}, author = {Amita Mishra and Sunita Soni}, doi = {10.1016/j.eswa.2025.129552}, journal = {Expert Systems with Applications}, month = {3}, pages = {129552}, shortjournal = {Expert Syst. Appl.}, title = {TriHAtt-BERT-DBiLSTM: Deep learning model for medical emergency prediction from the unstructured data}, volume = {298}, year = {2026}, }
• (2026). System-level integration of deep learning and computer vision for contact ring seal defect detection in semiconductor manufacturing. ESWA, 298, 129551. (www)

  @article{ , abstract = {The contact ring seals (CRSs) used in electroplating processes during semiconductor manufacturing are susceptible to degradation through chemical etching, electrochemical dissolution, and mechanical wear mechanisms. Despite the implementation of state-of-the-art surface treatment and coating technologies to mitigate CRS corrosion, manual intervention remains frequently required to address this problem. Conventional static defect detection systems for CRSs rely on predefined regions of interest (ROIs) and threshold-based defect area calculations, with surface anomalies identified by comparing the percentage of defective areas within these ROIs. However, this approach exhibits detection failures for millimeter-scale defects, low-contrast anomalies, and geometrically irregular patterns, especially under complex or dynamic environmental conditions. To address these systematic detection failures, we developed a dynamic defect detection system for CRSs by integrating artificial intelligence and traditional computer vision algorithms, achieving a 5.2x improvement in defect detection sensitivity. This system achieved detection accuracy and recall values of over 99 % as well as a response time of 1.43 s average latency, thereby demonstrating a substantial performance improvement compared to a static system, which achieved a recall rate of 18.9 % on the adopted dataset. The system satisfies real-time processing requirements while substantially reducing the need for manual intervention in defect detection and increases production efficiency. Finally, the experimental results of this study indicated that the postprocessing approaches used in the developed system enabled it to flexibly adapt to the different requirements of various production environments.}, archive = {J_ESWA}, author = {Ting-Han Chen and Hsin-Hung Chou and Shuang Zou and Yu-Han Chen and Sun-Yuan Hsieh}, doi = {10.1016/j.eswa.2025.129551}, journal = {Expert Systems with Applications}, month = {3}, pages = {129551}, shortjournal = {Expert Syst. Appl.}, title = {System-level integration of deep learning and computer vision for contact ring seal defect detection in semiconductor manufacturing}, volume = {298}, year = {2026}, }
• (2026). Complementary information-guided interactive fusion network for HSI and LiDAR data joint classification. ESWA, 298, 129549. (www)

  @article{ , abstract = {The classification of land remote sensing using single-modal data has reached a bottleneck, which has spurred significant interest in the joint utilization of multimodal remote sensing data to enhance classification performance. However, existing methods exhibit limitations in extracting intricate local and global features. Furthermore, achieving effective information interaction and deep fusion between multimodal datasets remains an unresolved challenge. To address these issues, we propose a Complementary Information-Guided Interactive Fusion Network (CIGIF-Net) for the classification of hyperspectral image (HSI) and light detection and ranging (LiDAR) data. The core idea of our approach leverages the capability of Convolutional Neural Networks (CNNs) to extract local spatial features while utilizing the strengths of Transformers in modeling long-range dependencies. Furthermore, our method facilitates deep fusion by designing mechanisms for the interactive integration of multimodal local spatial features, complemented by guidance from multimodal data during long-range dependency modeling, thereby improving overall classification performance. Specifically, CIGIF-Net incorporates multiscale feature learning, interactive feature fusion, and the complementary information-guided attention mechanism. Initially, CNNs are used to learn multiscale local spatial features. Subsequently, we perform an interactive fusion of multimodal spatial information based on channel attention techniques. Finally, the complementary information-guided attention mechanism dynamically utilizes complementary insights to inform deeper attention distributions, which guide global feature construction and enable efficient information aggregation. This methodology allows for the comprehensive extraction and synergistic utilization of complementary information across multimodal datasets. Extensive experiments conducted on three widely recognized HSI and LiDAR datasets demonstrate that the proposed CIGIF-Net achieves superior classification performance.}, archive = {J_ESWA}, author = {Shufang Xu and Qiyuan Xue and Zhonghao Chen and Shuyu Fei and Hongmin Gao}, doi = {10.1016/j.eswa.2025.129549}, journal = {Expert Systems with Applications}, month = {3}, pages = {129549}, shortjournal = {Expert Syst. Appl.}, title = {Complementary information-guided interactive fusion network for HSI and LiDAR data joint classification}, volume = {298}, year = {2026}, }
• (2026). A decision support system for open-pit mining optimisation with dynamic uncertainty and GPU-based parallel repair approach. ESWA, 298, 129544. (www)

  @article{ , abstract = {This paper presents an advanced Decision Support System (DSS) for long-term open-pit mine planning that integrates established optimization techniques—Large Neighborhood Search (LNS), Simulated Annealing (SA), and Dantzig-Wolfe decomposition—within a novel GPU-accelerated framework addressing geological uncertainty and computational complexity. The key methodological contributions include dynamic uncertainty modelling with time-dependent factors capturing geological confidence degradation and GPU-parallelized evaluation architecture enabling industrial-scale mine planning. Validation using 50,000 blocks across 10 geological scenarios demonstrates robust economic performance, achieving mean NPV of $1.514 billion with limited variability (standard deviation $16 million). The GPU-parallelized architecture achieves 29.6 % average computational speedup with peaks of 37 % compared to CPU implementations, enabling concurrent evaluation of 262,144 mining scenarios. Risk analysis reveals P90 Value-at-Risk of $1.488 billion, indicating strong downside protection. The system maintains profit margins exceeding 95 % across all scenarios with cumulative cash flow reaching $1.789.4 million by period 6. Narrow risk envelopes (P10-P90 spread <$60 M) demonstrate robust performance under uncertainty, providing mining companies with practical tools for risk-informed strategic decision-making.}, archive = {J_ESWA}, author = {Iman Rahimi}, doi = {10.1016/j.eswa.2025.129544}, journal = {Expert Systems with Applications}, month = {3}, pages = {129544}, shortjournal = {Expert Syst. Appl.}, title = {A decision support system for open-pit mining optimisation with dynamic uncertainty and GPU-based parallel repair approach}, volume = {298}, year = {2026}, }
• (2026). TriHID: Towards verifiable domain adaptation-based IoT intrusion detection in heterogeneous environment. ESWA, 298, 129543. (www)

  @article{ , abstract = {With the widespread deployment of IoT devices in daily applications, securing them against intrusions has become increasingly critical. Domain adaptation (DA)-based intrusion detection is a promising approach that transfers knowledge from a source domain to improve detection in a target IoT domain. However, effective DA methods must address various types of domain heterogeneity - such as differences in feature representation, intrusion distribution, and attack strategies. Existing intrusion detection datasets rarely consider these aspects, limiting their utility for evaluating heterogeneous DA approaches. To bridge this gap, we introduce TriHID , a new dataset specifically designed to capture heterogeneities from three perspectives. We evaluate four types of DA-based IoT intrusion detectors - multi-source, semi-supervised, unsupervised, and open-set on TriHID. Experimental results demonstrate that TriHID enables robust training and comprehensive evaluation of DA-based intrusion detection methods in heterogeneous IoT settings.}, archive = {J_ESWA}, author = {Jiashu Wu and Yang Wang}, doi = {10.1016/j.eswa.2025.129543}, journal = {Expert Systems with Applications}, month = {3}, pages = {129543}, shortjournal = {Expert Syst. Appl.}, title = {TriHID: Towards verifiable domain adaptation-based IoT intrusion detection in heterogeneous environment}, volume = {298}, year = {2026}, }
• (2026). Multi-objective dynamic flexible job shop scheduling using multi-head network-based deep reinforcement learning. ESWA, 298, 129542. (www)

  @article{ , abstract = {In modern manufacturing environments, job shop scheduling systems are characterized by heightened complexity and ever-changing dynamics, often involving multi-objective optimization and the need to accommodate unanticipated events like new job insertions and uncertain machine availability, underscores the necessity for effective real-time multi-objective scheduling approaches. Therefore, to tackle the multi-objective dynamic flexible job shop scheduling problem (MODFJSP) involving new job insertions, this paper introduces an online scheduling framework called multi-head deep Q network (MHDQN), designed to simultaneously minimize both total tardiness and total machine idle time. The core architecture of MHDQN framework is an innovative multi-head network agent based on Dueling deep Q network (Deuling DQN), consisting of a shared network layer and objective-specific network layers. The shared network layer extracts and transforms the input global state features layer by layer, generating a high-dimensional, semantically rich shared feature. This provides a unified input foundation for the objective-specific network layers, which are responsible for extracting the specialized information related to each objective from the shared features and calculating the corresponding Q -values, thereby enabling the parallel optimization of each objective. Six combined scheduling rules are developed to form the action set, each incorporating both job and machine selection. An improved multi-objective action selection strategy is proposed, incorporating inverse sigmoid ϵ decay and Q -value maximum absolute (max-abs) normalization to optimize decision-making. Additionally, a multi-head network training mechanism leveraging the Double deep Q network (Double DQN) architecture has been developed. Extensive computational experiments demonstrate that the MHDQN outperforms widely used traditional scheduling rules, multi-objective metaheuristic algorithms, and other reinforcement learning (RL) based scheduling methods, showing significant advantages and strong generalizability in multi-objective optimization tasks.}, archive = {J_ESWA}, author = {Kai Li and Bao Zheng and Liping Xu and Fulong Xie and Zhicheng Wang}, doi = {10.1016/j.eswa.2025.129542}, journal = {Expert Systems with Applications}, month = {3}, pages = {129542}, shortjournal = {Expert Syst. Appl.}, title = {Multi-objective dynamic flexible job shop scheduling using multi-head network-based deep reinforcement learning}, volume = {298}, year = {2026}, }
• (2026). A possibilistic programming approach in an integrated fuzzy periodic review model and clustering strategy for optimizing platelet supply chain. ESWA, 298, 129539. (www)

  @article{ , abstract = {Managing platelet supply chains poses significant challenges due to the product’s short shelf life, highly uncertain demand, and the critical nature of its medical use. Previous studies in the blood supply chain rely on fixed-order quantities and ignore collaborative inventory-sharing strategies, which can lead to either excessive waste or severe shortages. However, in many real-world situations, fixed order quantities are often insufficient to accommodate fluctuating demand, especially in healthcare systems. Moreover, existing distribution models in the literature often overlook the equitable allocation of services across hospitals, leading to disparities in access to critical healthcare resources. This study proposes a novel two-phase decision-making framework that integrates a fuzzy periodic review inventory model with a cluster-based reactive transshipment strategy to optimize platelet supply and distribution. In Phase I, a fuzzy periodic review model determines optimal order quantities under uncertain demand using a possibilistic chance-constrained programming approach. In Phase II, hospitals are clustered based on service levels, enabling equitable transshipment among facilities to reduce disparities and improve overall responsiveness. A real-world case study from Tehran province is used to evaluate the model’s effectiveness. Results show an approximate 6% reduction in total shortages and a 2% improvement in average service levels. The proposed framework offers actionable insights for healthcare managers aiming to enhance resilience, equity, and efficiency in critical medical supply chains.}, archive = {J_ESWA}, author = {Seyyed-Mahdi Hosseini-Motlagh and Mohammad Reza Ghatreh Samani and Hannaneh Kordhaghi}, doi = {10.1016/j.eswa.2025.129539}, journal = {Expert Systems with Applications}, month = {3}, pages = {129539}, shortjournal = {Expert Syst. Appl.}, title = {A possibilistic programming approach in an integrated fuzzy periodic review model and clustering strategy for optimizing platelet supply chain}, volume = {298}, year = {2026}, }
• (2026). Deep-insights guided evolutionary algorithm for optimization. ESWA, 298, 129538. (www)

  @article{ , abstract = {Evolutionary algorithms (EAs) are a class of optimization algorithms inspired by the theory of biological evolution. They solve optimization problems by emulating the processes of natural selection. EAs produce abundant data during evolution, which contains valuable information that reflects their evolutionary patterns. Effectively utilizing this information can enhance the algorithms’ effectiveness and efficiency. Deep learning excels at extracting knowledge from data. Inspired by this, we propose a novel insights-infused framework that utilizes deep neural networks to learn the evolutionary processes of EAs and extract useful synthesis insights from the evolutionary data. These synthesis insights not only guide the algorithm to evolve in a better direction on the original problems, but also improve its performance on new problems. The choice of neural networks is important. During pre-training, to reduce the inductive bias introduced by human prior knowledge, we design an MLP model to process the data. Additionally, we develop a variable-length encoding method to enable MLP networks to handle variable-length data. To verify the transfer evolution ability of synthesis insights, we devise a self-evolution strategy that fine-tunes the network using only the data generated by the algorithm itself, without introducing any external knowledge, when dealing with new problems. Experimental results demonstrate that the synthesis insights extracted from the CEC2014 dataset guide the algorithms to evolve in a better direction for the CEC2014 problems, and in addition enhance their performance on new problems like CEC2017, CEC2022 and the real-world optimization problems.}, archive = {J_ESWA}, author = {Kun Bian and Juntao Zhang and Hong Han and Jun Zhou and Yifei Sun and Shi Cheng}, doi = {10.1016/j.eswa.2025.129538}, journal = {Expert Systems with Applications}, month = {3}, pages = {129538}, shortjournal = {Expert Syst. Appl.}, title = {Deep-insights guided evolutionary algorithm for optimization}, volume = {298}, year = {2026}, }
• (2026). MambaGen: Efficient visual representation learning for automatic radiology report generation. ESWA, 298, 129537. (www)

  @article{ , abstract = {Radiology report generation focuses on producing comprehensive and clinically precise medical reports based on radiographic images, thereby improving medical efficiency and alleviating the burden on radiologists. Although existing deep learning methods have demonstrated superior performance, they are constrained by the local receptive field of convolutional neural networks and are inadequate for modeling long-range dependencies, making it challenging to detect critical lesion features in medical images. Recently, State Space Models (SSMs), particularly Mamba, have shown great potential in modeling long-range dependencies with linear computational complexity. Inspired by this, we propose MambaGen, the enhanced Mamba model specifically designed for radiology report generation tasks. Specifically, we design a Mamba-Visual Recalibration Module (MVRM), which utilizes a two-stage training strategy to effectively capture the efficient visual representation of medical images. This first stage combines convolutional layers with SSMs to model long-sequence dependencies and learn multi-level visual feature information. This second stage introduces local convolution and a channel attention mechanism to further recalibrate the local feature and mitigate channel redundancy. Comprehensive experiments on widely available datasets, such as IU X-Ray and MIMIC-CXR, demonstrate our model’s superior performance compared to existing methods, particularly with an improvement of 2.7 % on the BLEU-4 metric. The code is available at https://github.com/Eleanorhxd/MambaGen.git .}, archive = {J_ESWA}, author = {Xiaodi Hou and Xiaobo Li and Simiao Wang and Mingyu Lu and Hongfei Lin and Yijia Zhang}, doi = {10.1016/j.eswa.2025.129537}, journal = {Expert Systems with Applications}, month = {3}, pages = {129537}, shortjournal = {Expert Syst. Appl.}, title = {MambaGen: Efficient visual representation learning for automatic radiology report generation}, volume = {298}, year = {2026}, }
• (2026). Multistep diesel vehicle emissions forecasting with an efficient transformer enhanced by temporal-frequency fusion and covariate interaction. ESWA, 298, 129532. (www)

  @article{ , abstract = {Controlling nitrogen oxide (NO x ) emissions from diesel vehicles is a critical environmental challenge. Advanced SCR strategies depend on accurate multistep forecasting of NO x and ammonia (NH 3 ), but existing models often struggle with error accumulation and the non-stationary dynamics of emissions data. In this work, we first establish a new benchmark for this task, confirming that handling data non-stationarity is essential for robust prediction. Building on this insight, we propose FiTformer, a novel Transformer architecture that adopts an encoder-only framework to jointly forecast both NO x and NH 3 concentrations, where we introduce Intra-series Temporal-Frequency Fusion mechanism to capture intrinsic emissions dynamics and Inter-series Covariate Interaction mechanism to model external influences. Validated on real-world engine data, FiTformer consistently outperforms baseline models across all evaluated prediction horizons, with up to 44.1 % MAE and 36.9 % SMAPE reductions in 24-step NO x prediction and similarly strong gains for NH 3 prediction, compared to the state-of-the-art baseline TimesNet. Its high computational efficiency (0.30G MACs and 8.3 ms/iter) along with robust generalization and high resilience to data imperfections, underscores its suitability for real-time embedded SCR control, enabling more effective strategies for NO x reduction and NH 3 slip minimization.}, archive = {J_ESWA}, author = {Yuhan Luo and Yujun Zhang and Ying He and Kun You and Wei Huang and Wenqing Liu and Hao Xie}, doi = {10.1016/j.eswa.2025.129532}, journal = {Expert Systems with Applications}, month = {3}, pages = {129532}, shortjournal = {Expert Syst. Appl.}, title = {Multistep diesel vehicle emissions forecasting with an efficient transformer enhanced by temporal-frequency fusion and covariate interaction}, volume = {298}, year = {2026}, }
• (2026). An EEG-based dual-stream spatial-spectral-temporal large model for self-limited epilepsy with centrotemporal spikes. ESWA, 298, 129530. (www)

  @article{ , abstract = {Self-limited epilepsy with centrotemporal spikes (SeLECTS) is the most common form of focal epilepsy in childhood, accounting for 20–25 % of all childhood epilepsy cases and may be associated with cognitive dysfunction and behavioral issues. Accurate detection and assessment of epileptic discharges in EEG signals, particularly the spike-wave index (SWI), are crucial for timely intervention and treatment. Manual analysis of EEG data is labor-intensive and prone to errors, underscoring the need for automated methods. In the present study, we propose a novel D ual-Str e am Sp a tial- S pectral- T emporal L arge model (DeaSTL) that leverages a large-scale EEG architecture to effectively capture the multidimensional characteristics of EEG signals associated with SeLECTS syndrome. Our model integrates multi-view temporal representations and spatial-spectral representations through a dual-stream approach, enhancing the learning of complex patterns in EEG data. We introduce the S JTU Se L ECTS E EG D ataset (SLED), a comprehensive EEG dataset from 212 patients diagnosed with SeLECTS, including annotations for abnormal discharge detection, wake-sleep period classification, and SWI estimation. Addressing the previously unexplored problem of SWI prediction, we provide a novel method for quantifying the severity of epileptic discharges during sleep. Extensive experiments demonstrate that our DeaSTL model significantly outperforms several state-of-the-art methods across multiple tasks, showcasing its potential for clinical application in assisting diagnosis and treatment planning.}, archive = {J_ESWA}, author = {Lin Zhang and Yun Ren and Fang Yuan and Xuqin Chen and Shikui Tu and Lei Xu}, doi = {10.1016/j.eswa.2025.129530}, journal = {Expert Systems with Applications}, month = {3}, pages = {129530}, shortjournal = {Expert Syst. Appl.}, title = {An EEG-based dual-stream spatial-spectral-temporal large model for self-limited epilepsy with centrotemporal spikes}, volume = {298}, year = {2026}, }
• (2026). Dynamic learning of sample ambiguity-driven sample weighting for medical image classification. ESWA, 298, 129527. (www)

  @article{ , abstract = {Deep neural networks have delivered impressive results in medical image classification tasks. However, their performance is still challenging in medical scenarios with limited data, where training set biases such as label noise or class imbalance impede model learning. Dynamic learning based sample weighting achieves adaptive adjustment of sample importance through learnable weight functions and shows great potential in improving model robustness. Nevertheless, existing methods directly employ model states such as loss value or training accuracy to evaluate sample importance, ignoring the role of ambiguous samples in model optimization. This limitation hinders the performance of dynamic learning based sample weighting in medical image classification. In this paper, we propose a new sample weighting approach based on sample ambiguity and dynamic learning for improving medical image classification, named DLSA-SW. We introduce a dual-space sample ambiguity method by evaluating the category proximity in the feature space and the prediction confidence in the label space. Subsequently, to dynamically calculate sample weights according to sample ambiguity, a learnable sample weighting network is developed to adaptively adjust the weights during training to guide the task model. DLSA-SW performs alternate optimization to enable mutual adaptation of the sample weighting network and the task network. We evaluate the effectiveness of our approach on three medical image classification benchmarks: PatchCamelyon for lymph node histopathology classification, ISIC 2020 for skin lesion classification, and MTC for medullary thyroid cancer classification. DLSA-SW outperforms existing state-of-the-art sample weighting methods on all three datasets and yields substantial improvements over methods without sample weighting. These results demonstrate the robustness and practical applicability of our approach in clinical diagnostic tasks.}, archive = {J_ESWA}, author = {Guanxiu Yi and Xiabi Liu and Ling Ma and Mengqiao Han and Lijuan Niu}, doi = {10.1016/j.eswa.2025.129527}, journal = {Expert Systems with Applications}, month = {3}, pages = {129527}, shortjournal = {Expert Syst. Appl.}, title = {Dynamic learning of sample ambiguity-driven sample weighting for medical image classification}, volume = {298}, year = {2026}, }
• (2026). LST-rPPG: A long-range spatio-temporal model for high-accuracy heart rate variability measurement. ESWA, 298, 129526. (www)

  @article{ , abstract = {Remote photoplethysmography (rPPG) is a non-contact manner of measuring heart rate variability (HRV) by deriving blood volume pulse (BVP) signals from facial videos. The performance of rPPG-based HRV measurement is challenging due to short-range noises (e.g., head movements) suppression and sufficient-duration BVP signal generation. Recent Transformer-based rPPG methods have shown advantages of global spatio-temporal feature modeling in eliminating noise and recovering high-quality BVP signals. However, these methods often face significant computational and memory constraints, limiting duration scalability of the generated BVP signals that might decrease HRV measurement performance. To address the above issue, this paper proposes a duration-scalable Transformer-based rPPG method, capable of global Long-range Spatio-Temporal modelling (LST), termed LST-rPPG, to generate high-quality BVP signals with a much longer duration. On the one hand, by employing spatial and temporal encoders, the original image-based rPPG issue is converted to a time-series problem. Besides, a sparse computation mechanism is integrated into the temporal encoder. This combination allows LST-rPPG to recover flexible-duration BVP signals, supporting continuous modeling of segments beyond 30 s with low computation and memory overhead. On the other hand, a dynamic loss function with stringent temporal constraints is designed to guarantee the quality of the generated BVP signals. Comprehensive experiments are performed on two public datasets, PURE and UBFC-RPPG, and the results demonstrate the feasibility of LST-rPPG for generating high-quality BVP signals with a much longer duration while requiring substantially fewer computational resources. Besides, LST-rPPG achieves at least the second-best results during all experiments.}, archive = {J_ESWA}, author = {Jiajie Li and Juan Cheng and Rencheng Song and Yu Liu}, doi = {10.1016/j.eswa.2025.129526}, journal = {Expert Systems with Applications}, month = {3}, pages = {129526}, shortjournal = {Expert Syst. Appl.}, title = {LST-rPPG: A long-range spatio-temporal model for high-accuracy heart rate variability measurement}, volume = {298}, year = {2026}, }
• (2026). A novel scheme integrating graph-based analysis and opposition-based learning for S-box optimization by population-based metaheuristics. ESWA, 298, 129524. (www)

  @article{ , abstract = {The non-linearity of symmetric block cryptographic algorithms, which is crucial for resisting linear and differential cryptanalysis, depends on the design of substitution boxes. This work proposes a novel scheme, named GOM, combining opposition-based learning with graph-based representation to be integrated into three population-based metaheuristics. This scheme is applied to address the 8 × 8 S-box design problem. The GOM-enhanced metaheuristics improve population diversity and convergence, achieving a non-linearity of 112. Hardware simulations indicate that GOM-based designs require similar resources to AES, while side-channel evaluations confirm their resilience against power analysis attacks. Scalability is supported by successful results on 10 × 10 and 12 × 12 S-boxes. The design of GOM, leveraging generalizable components such as opposition-based learning and graph-based representations, suggests its potential applicability to other population-based metaheuristics and optimization problems.}, archive = {J_ESWA}, author = {Francisco González and Ricardo Soto and José M. Lanza-Gutiérrez and Broderick Crawford}, doi = {10.1016/j.eswa.2025.129524}, journal = {Expert Systems with Applications}, month = {3}, pages = {129524}, shortjournal = {Expert Syst. Appl.}, title = {A novel scheme integrating graph-based analysis and opposition-based learning for S-box optimization by population-based metaheuristics}, volume = {298}, year = {2026}, }
• (2026). Functional consistency of LLM code embeddings: A self-evolving data synthesis framework for benchmarking. ESWA, 298, 129523. (www)

  @article{ , abstract = {Embedding models have demonstrated strong performance in tasks like clustering, retrieval, and feature extraction while offering computational advantages over generative models and cross-encoders. Benchmarks such as MTEB have shown that text embeddings from large language models (LLMs) capture rich semantic information, but their ability to reflect code-level functional semantics remains unclear. Existing studies largely focus on code clone detection, which emphasizes syntactic similarity and overlooks functional understanding. In this paper, we focus on the functional consistency of LLM code embeddings, which determines if two code snippets perform the same function regardless of syntactic differences. We propose a novel data synthesis framework called Functionality-Oriented Code Self-Evolution to construct diverse and challenging benchmarks. Specifically, we define code examples across four semantic and syntactic categories and find that existing datasets predominantly capture syntactic properties. Our framework generates four unique variations from a single code instance, providing a broader spectrum of code examples that better reflect functional differences. Extensive experiments on three downstream tasks-code clone detection, code functional consistency identification, and code retrieval-demonstrate that embedding models significantly improve their performance when trained on our evolved datasets. These results highlight the effectiveness and generalization of our data synthesis framework, advancing the functional understanding of code.}, archive = {J_ESWA}, author = {Zhuohao Li and Wenqing Chen and Jianxing Yu and Zhichao Lu}, doi = {10.1016/j.eswa.2025.129523}, journal = {Expert Systems with Applications}, month = {3}, pages = {129523}, shortjournal = {Expert Syst. Appl.}, title = {Functional consistency of LLM code embeddings: A self-evolving data synthesis framework for benchmarking}, volume = {298}, year = {2026}, }
• (2026). Semisupervised graph U-net with G-ConvLSTM for hyperspectral image classification. ESWA, 298, 129522. (www)

  @article{ , abstract = {Graph convolutional networks (GCNs) have demonstrated the effectiveness for hyperspectral image (HSI) classification, but still face challenges, such as insufficient exploitation of data structure information, limited labeled samples, and high susceptibility to noise and outliers. To address these issues, a semisupervised graph U-Net with graph convolutional long short-term memory is proposed for HSI classification, abbreviated as SSGU-Net. Specifically, we design a novel graph convolutional long short-term memory feature extractor to learn discriminative spatial-spectral joint features by simultaneously modeling the correlations in the spatial and spectral domains. Then, we develop a semisupervised graph U-Net with mutually inverse operation of the graph pooling and the graph unpooling modules which uses both labeled samples and unlabeled samples to train a well-parameterized network for HSI classification. In particular, to suppress the effects of noise and outliers, the graph pooling module is designed to selectively retain discriminative samples and fully learn the optimal correlation between these retained samples. Meanwhile, the graph unpooling module employs the local spatial context to reconstruct the reduced samples, thus restoring the pooled data to its original scale for classification task. Extensive experiments show the effectiveness of the proposed method, achieving overall accuracy gains of 5.22 %, 1.58 %, and 1.57 % over the state-of-the-art competitors on the Indian Pines, University of Pavia, and Houston datasets, respectively.}, archive = {J_ESWA}, author = {Jin-Yu Yang and Heng-Chao Li and Xin-Ru Feng and Feng Gao and Qian Du and Antonio Plaza}, doi = {10.1016/j.eswa.2025.129522}, journal = {Expert Systems with Applications}, month = {3}, pages = {129522}, shortjournal = {Expert Syst. Appl.}, title = {Semisupervised graph U-net with G-ConvLSTM for hyperspectral image classification}, volume = {298}, year = {2026}, }
• (2026). Chat with one voice: Mitigating semantic disparity across languages for multilingual open-domain dialogue response generation systems. ESWA, 298, 129521. (www)

  @article{ , abstract = {With the development of the Internet, building a multilingual dialogue generation system to attract more users while reducing costs in the global market has become increasingly important. However, current end-to-end multilingual approaches often face semantic disparity issues across languages. When given parallel queries with the same semantics but in different languages, the generated responses may vary in meaning across languages, which may greatly affect the stability and reliability of multilingual systems in different language scenarios. We attribute this issue to open-domain/model uncertainty and language differences. To mitigate this issue, we first propose a novel Anchor-based Semantic Constraint (ASC) designed to reduce semantic disparity across languages for Encoder-Decoder Transformers. ASC employs language-independent anchor signal to guide the behaviors in both the encoder and decoder, thereby reducing uncertainty. Additionally, ASC incorporates a two-stage tuning process to further minimize the impact of language differences by ensuring the encoder remains language-independent. Extensive experiments and in-depth analyses conducted on XDailyDialog demonstrate that ASC can effectively mitigate semantic disparity across languages and will not compromise dialogue response quality like the previous related baselines.}, archive = {J_ESWA}, author = {Sixing Wu and Jiahao Chen and Jiong Yu and Wei Zhou}, doi = {10.1016/j.eswa.2025.129521}, journal = {Expert Systems with Applications}, month = {3}, pages = {129521}, shortjournal = {Expert Syst. Appl.}, title = {Chat with one voice: Mitigating semantic disparity across languages for multilingual open-domain dialogue response generation systems}, volume = {298}, year = {2026}, }
• (2026). Towards CPU performance prediction: New challenge benchmark dataset and novel approach. ESWA, 298, 129520. (www)

  @article{ , abstract = {CPU performance prediction based on hardware characteristics is crucial for system design and resource management. However, this field faces two major challenges. First, collecting real-world data is challenging due to the diversity of CPU products and the specialized nature of hardware characteristics. This field lacks a standard dataset with unified hardware characteristics, wide data coverage, and comprehensive benchmarks. Second, existing methods based on hardware simulation models or machine learning exhibit notable shortcomings, such as lengthy simulation test cycles and low prediction accuracy. To bridge these gaps, we first collect, preprocess, and standardize historical data from the 4th Generation Intel® Xeon® Scalable Processors across multiple benchmark suites to create a new dataset, named PerfCastDB. Subsequently, we design a deep learning based model called Nova CPU Performance Predictor (NCPP) as the baseline for this new dataset. The NCPP network is designed based on group attention mechanism. It effectively quantifies the implicit relationships between hardware characteristics within and across groups and comprehensively models the impact of various hardware characteristics on CPU performance prediction. We conduct comparative experiments using the proposed PerfCastDB dataset. Compared to existing approaches, NCPP achieves superior evaluation results, demonstrating its effectiveness. Furthermore, we have open-sourced part of the dataset and the NCPP network code to facilitate subsequent research. The resources can be accessed at https://github.com/xiaoman-liu/NCPP .}, archive = {J_ESWA}, author = {Xiaoman Liu}, doi = {10.1016/j.eswa.2025.129520}, journal = {Expert Systems with Applications}, month = {3}, pages = {129520}, shortjournal = {Expert Syst. Appl.}, title = {Towards CPU performance prediction: New challenge benchmark dataset and novel approach}, volume = {298}, year = {2026}, }
• (2026). Glyph graph isomorphism network for structure recognition of oracle bone inscription. ESWA, 298, 129519. (www)

  @article{ , abstract = {Structure recognition of oracle bone inscription glyphs plays an important role in studying the evolutionary process of oracle bone inscriptions and the history of the Shang Dynasty. Currently, most methods have decomposed oracle bone inscription glyphs into multilevel features, which are used to recognize hierarchical feature fusion. This strategy cannot recognize the primitive internal structures of keypoints, strokes, and components. Moreover, mainstream graph neural networks cannot fully utilize the rich structural information of oracle bone inscription glyphs, resulting in their inability to meet the needs of structure recognition. So we have developed a graph structure recognition method to implement structure recognition of oracle bone inscription glyphs. A graph extraction method is given to get the structure of oracle bone inscription glyphs; each graph structure’s representation vector can be learned by a glyph graph isomorphism network, which is developed to recognize the graph of oracle bone inscription glyphs to enhance the discriminability representation of the structure. Our model has achieved advanced results across structure recognition experiments in the HWOBC dataset and the Oracle-50K dataset.}, archive = {J_ESWA}, author = {Zhan Zhang and Hanbin Liu and Xingkun Zhang and Yiyuan Wang and Feng Gao and An Guo and Han Zhang and Qingju Jiao and Bang Li and Yongge Liu}, doi = {10.1016/j.eswa.2025.129519}, journal = {Expert Systems with Applications}, month = {3}, pages = {129519}, shortjournal = {Expert Syst. Appl.}, title = {Glyph graph isomorphism network for structure recognition of oracle bone inscription}, volume = {298}, year = {2026}, }
• (2026). An integrated decision-making framework to evaluate the route alternatives in overweight/oversize transportation. ESWA, 298, 129516. (www)

  @article{ , abstract = {Overweight and oversized transport (O&OT) has become one of the most critical elements of project logistics, driven by advancements in transportation and lifting technologies that now allow high-volume loads to be moved across long distances. This type of transportation operation, also called abnormal transportation, is greatly affected by technical factors such as the weight and geometry of the load, road surface, axle load limitations, slope, and ground strength, as well as external variables such as weather conditions, traffic density, and legal regulations. In planning and operational processes, Decision-Makers (DMs) and practitioners who plan and execute operations without adequately considering these factors and variables can lead to delays in operations, serious risks, and loss of productivity. This research proposes a flexible decision support model that integrates Step-wise Weight Assessment Ratio Analysis (SWARA) and Logarithmic Percentage Change-driven Objective Weighting (LOPCOW), and a ranking technique; i.e., Mixed Aggregation by Comprehensive Normalization Technique (MACONT) techniques to address the decision problems related to route selection, one of the most critical problems in transporting heavy and bulky loads, and to produce reasonable solutions. The proposed model significantly reduces information losses by processing subjective and objective information and integrating subjective (SWARA) and objective (LOPCOW) methods. Unlike traditional ranking approaches, the MACONT method combines three different normalization techniques to determine the ranking performance of alternatives. In this way, it provides more reliable and accurate results by reducing the deviations of the results provided by the single normalization technique. In addition, it shows each alternative’s good and bad performance compared to the others and is more convincing about the results obtained. According to the results obtained by applying the proposed model, fuel consumption (0.096) is determined as the most effective and critical factor in selecting the route on which heavy and bulky loads will be transported. In this context, choosing routes that allow lower fuel consumption can contribute to reducing carbon emissions and external costs arising from transportation. The extensive robustness and validation check to test the proposed model prove that the proposed model is a reliable, robust, and practical decision-making tool for making reasonable and rational decisions in O&OT.}, archive = {J_ESWA}, author = {Ömer Faruk Görçün and Pradip Kundu and Hande Küçükönder and Gürkan Doğan and Erfan Babaee Tirkolaee}, doi = {10.1016/j.eswa.2025.129516}, journal = {Expert Systems with Applications}, month = {3}, pages = {129516}, shortjournal = {Expert Syst. Appl.}, title = {An integrated decision-making framework to evaluate the route alternatives in overweight/oversize transportation}, volume = {298}, year = {2026}, }
• (2026). EvoMapX: An explainable framework for metaheuristic optimization algorithms. ESWA, 298, 129514. (www)

  @article{ , abstract = {Population-based optimization algorithms (POAs) are widely adopted solutions for NP-hard and complex high-dimensional optimization problems. However, their internal dynamics often remain opaque, limiting trust and insight into how solutions evolve. This paper introduces EvoMapX, a novel explainable framework designed to interpret the internal dynamics of population-based optimization algorithms. EvoMapX includes three interpretable structures to visualize evolutionary optimization dynamics: the Operator Attribution Matrix (OAM) quantifies the contribution of specific operators over iterations; the Population Evolution Graph (PEG) traces the ancestry and transformation of candidate solutions; and the Convergence Driver Score (CDS) identifies which operators drive convergence, helping interpret why the algorithm improved. EvoMapX was evaluated across four POAs on the CEC 2021 test suite in order to demonstrate how it reveals meaningful textual and graphical insights into algorithm behavior. EvoMapX paves the way for interpretable metaheuristic optimization. The source code of EvoMapX is available at https://www.github.com/Bilal20252025/EvoMapX}, archive = {J_ESWA}, author = {Bilal H. Abed-Alguni}, doi = {10.1016/j.eswa.2025.129514}, journal = {Expert Systems with Applications}, month = {3}, pages = {129514}, shortjournal = {Expert Syst. Appl.}, title = {EvoMapX: An explainable framework for metaheuristic optimization algorithms}, volume = {298}, year = {2026}, }
• (2026). BDOS: An orthogonal sum based on belief of permutation events and element distances in random permutation set. ESWA, 298, 129513. (www)

  @article{ , abstract = {The random permutation set (RPS) extends Dempster-Shafer evidence theory by incorporating event order information, providing a powerful framework for modeling uncertainty. However, existing orthogonal sum methods within the RPS framework may encounter loss of order information and counterintuitive belief distribution during permutation event fusion. To address these two issues, this paper proposes a new method termed belief-distance-based orthogonal sum (BDOS). BDOS operates through three core mechanisms: order-information preservation via mathematical constructs like order-space and inverse mapping; belief-value weighting that prioritizes events with high belief mass for rational outcomes; and element-distance weighting that incorporates dissimilarity among permutations to improve ordinal accuracy. Numerical examples validate the effectiveness of BDOS in permutation event fusion, with comparative results demonstrating its advantages in order retention and belief distribution. Furthermore, BDOS is applied to threat assessment, illustrating its rationality and effectiveness in handling uncertainty and threat ranking.}, archive = {J_ESWA}, author = {Xiaoyan Su and Xu Chen and Hong Qian and Cheng Jiang}, doi = {10.1016/j.eswa.2025.129513}, journal = {Expert Systems with Applications}, month = {3}, pages = {129513}, shortjournal = {Expert Syst. Appl.}, title = {BDOS: An orthogonal sum based on belief of permutation events and element distances in random permutation set}, volume = {298}, year = {2026}, }
• (2026). Fine-grained similarity retrieval of lesion areas in lung CT images based on visual similarity matching of image blocks. ESWA, 298, 129510. (www)

  @article{ , abstract = {Background and Objective With the rapid growth in the number of medical images, the need for content- based medical image retrieval (CBMIR) in clinical aid diagnosis is becoming increasingly important. Most current content-based CT image similarity retrieval methods use the entire CT image, ignoring the fact that the localized lesion region is the main target of similarity retrieval; Methods To address this issue, the paper proposes a fine-grained similarity retrieval method for lung CT images based on image block( IB ) similarity matching, taking lung CT images as an example. In this method, two enabling techniques are introduced: 1) a hybrid Convolution and Vision Transformer Model(CVTM) that effectively captures both local texture and global context features of lesion regions; 2) the iDS high-dimensional index designed to accelerate retrieval among IB ; Results With the aid of these techniques, fine-grained similarity retrieval optimization of lung CT images can be achieved, which facilitates more accurate lesion-level comparison and supports clinical decision-making; Conclusions Extensive experiments are conducted to indicate that the proposed fine-grained similarity retrieval method achieves excellent performance, with a mAP of 91.33%. Meanwhile, the retrieval efficiency of the iDS high-dimensional index is about 150% higher than that of sequential retrieval, especially when the retrieval radius is large and the database size is substantial.}, archive = {J_ESWA}, author = {Yi Zhuang and Jiayu Zhang and Yujia Ge and Nan Jiang}, doi = {10.1016/j.eswa.2025.129510}, journal = {Expert Systems with Applications}, month = {3}, pages = {129510}, shortjournal = {Expert Syst. Appl.}, title = {Fine-grained similarity retrieval of lesion areas in lung CT images based on visual similarity matching of image blocks}, volume = {298}, year = {2026}, }
• (2026). Traffic prediction using an active causality recurrent graph convolutional network. ESWA, 298, 129506. (www)

  @article{ , abstract = {The quality of our daily lives is significantly influenced by traffic conditions, highlighting the importance of incorporating complex spatiotemporal dependencies in interconnected traffic data for effective prediction. Although recent advancements have demonstrated prediction accuracy using graph convolutional networks, their depends heavily on the accuracy of the graph structures that represent the spatial relationships within the traffic network. To address this challenge, we introduce a novel approach to traffic prediction, the Active Causal Recurrent Graph Convolution Network (ACRGCN), as shown in Fig. 2. ACRGCN offers a new framework that effectively integrates a causal-embedded approach for traffic prediction, leveraging both structural and feature information from correlated traffic time series. Additionally, it incorporates a time-varying dynamic Bayesian network to capture the intricate spatiotemporal topology of traffic data. The model extracts spatiotemporal dependencies from traffic signals using the Active Causality Graph Recurrent Module (ACGRM), while efficiently modeling nonlinear traffic propagation patterns. Furthermore, ACRGCN employs a deep learning-based module that functions as a hyper-network, progressively generating dynamic causal graphs. Finally, extensive experiments on multiple real-world traffic graph datasets validate ACRGCN, and the results demonstrate its superiority over state-of-the-art method}, archive = {J_ESWA}, author = {Jinde Zhu and Junhao Yuan and Fulan Ye and Trong-The Nguyen and Ruoxi Wang and Wu Zeng and Chien-Chun Liu}, doi = {10.1016/j.eswa.2025.129506}, journal = {Expert Systems with Applications}, month = {3}, pages = {129506}, shortjournal = {Expert Syst. Appl.}, title = {Traffic prediction using an active causality recurrent graph convolutional network}, volume = {298}, year = {2026}, }
• (2026). Shielding federated learning: Mitigating label transferability against poisoning attacks in cloud-edge-client system. ESWA, 298, 129502. (www)

  @article{ , abstract = {Hierarchical computing architecture of cloud-edge-client (CEC) formed with the combination of cloud computing and edge computing can provide processing, storage and low-latency services close to end devices. To protect data privacy, federated learning (FL), as a novel intelligent edge computing framework with localized training mechanisms, has been integrated into edge computing to form a system called CEC-FL and is widely studied. However, they are susceptible to potential poisoning attacks. Existing poisoning attack methods are mostly explored by performing malicious operations on training samples or labels directly and implementing corresponding defense strategies: they are designed to ignore the label transferability and diverse attack environments and are not work against stealthy security threats, mainly because they do not take into account the inherent vulnerabilities of the attack environment. Yet few general defense schemes have been developed. In response to the above vulnerabilities, in this work, we explore a B arycenter Po isoning method with L abel T ransferability (BPoLT) initiated by malicious attackers, resulting in a dynamic attack capability on the CEC-FL system. To address poisoning attacks, we provide a two-phase defense algorithm Res isting L abel T ransferability Pois oning called ResLT-Pois to distinguish malicious attackers from benign participants. Extensive experimental results demonstrate that our scheme is feasible and effective in dealing with the vulnerability of the CEC-FL system.}, archive = {J_ESWA}, author = {Yaru Zhao and Yihao Cao and Jianbiao Zhang and Zhaoqian Zhang and Weiru Wang}, doi = {10.1016/j.eswa.2025.129502}, journal = {Expert Systems with Applications}, month = {3}, pages = {129502}, shortjournal = {Expert Syst. Appl.}, title = {Shielding federated learning: Mitigating label transferability against poisoning attacks in cloud-edge-client system}, volume = {298}, year = {2026}, }
• (2026). SAMForecast: A hybrid model of self-attention and mamba with adaptive wavelet transform for time series forecasting. ESWA, 298, 129498. (www)

  @article{ , abstract = {Time series forecasting faces significant challenges due to the variability and complexity of real-world data. Traditional methods often require manual adjustments of wavelet transform parameters, which are labor-intensive and prone to over-fitting or inadequate feature extraction. To address these limitations, this study proposes SAMForecast, a novel hybrid model that integrates Adaptive Wavelet Transform, self-attention mechanisms, and the selective state space model. We introduce an Adaptive Wavelet Block that dynamically adjusts decomposition levels and basis functions using a Mixture of Experts network and lifting scheme, eliminating the need for manual parameter tuning. Furthermore, the model deeply integrates the attention mechanism of the Transformer architecture, leveraging its advantages in capturing complex dependencies to identify correlations between time series data. By combining self-attention with Mamba, SAMForecast effectively captures both global dependencies and local key features in time series, enhancing robustness against noise and redundant information. SAMForecast demonstrates promising performance in multivariate time series forecasting tasks, showcasing an average 2 % performance improvement compared to existing models across datasets in energy, transportation, and other fields. The code is available at https://github.com/Kiki-V/SAMForecast-main .}, archive = {J_ESWA}, author = {Dunlu Peng and Qiqi Lin}, doi = {10.1016/j.eswa.2025.129498}, journal = {Expert Systems with Applications}, month = {3}, pages = {129498}, shortjournal = {Expert Syst. Appl.}, title = {SAMForecast: A hybrid model of self-attention and mamba with adaptive wavelet transform for time series forecasting}, volume = {298}, year = {2026}, }
• (2026). An adaptive hypergraph-based convolution network with dual spatiotemporal attention for PM2.5 forecasting. ESWA, 298, 129497. (www)

  @article{ , abstract = {Accurate PM2.5 prediction is crucial for effective environmental management and public health protection, yet current models show limited dynamic adaptability to complicated air pollution scenarios. Robust models are essential to support timely interventions in response to sudden pollution events or rapidly changing air quality patterns. However, existing models predominantly rely on predefined graph structures and pairwise spatial relationships, limiting their ability to capture the complex and dynamic interactions inherent in PM2.5 pollution. Furthermore, such models often assume equal contributions from neighboring nodes, neglecting heterogeneity and compromising predictive accuracy. To address these limitations, we propose an Adaptive Hypergraph-based Convolution Network with a Dual Spatiotemporal Attention mechanism (AHCN-DA) for PM2.5 forecasting. This framework leverages representation learning and hypergraph structures to capture and integrate pairwise as well as higher-order spatial interactions, producing richer spatial-feature representations. The dual spatiotemporal attention mechanism dynamically assigns time-varying weights to neighboring nodes based on their relevance to target nodes, effectively mitigating the impact of irrelevant inputs. Additionally, AHCN-DA integrates a dilated convolution network with multi-scale kernels to capture temporal patterns effectively across varying scales. Extensive experiments on the 2023 China National Air Quality Dataset show significant improvements in predictive accuracy, particularly in enhancing the proportion of high-precision monitoring stations, with an R 2 of 0.9224, outperforming baseline models. Our findings underscore the effectiveness of AHCN-DA in enhancing prediction accuracy under complex pollution response patterns, contributing to more informed decision-making in environmental management.}, archive = {J_ESWA}, author = {Haipeng Gao and Chonghui Qian and Yang Su and Wei Zhang and Hengjun Huang}, doi = {10.1016/j.eswa.2025.129497}, journal = {Expert Systems with Applications}, month = {3}, pages = {129497}, shortjournal = {Expert Syst. Appl.}, title = {An adaptive hypergraph-based convolution network with dual spatiotemporal attention for PM2.5 forecasting}, volume = {298}, year = {2026}, }
• (2026). Solving equilibrium for adversarial team games utilizing fictitious team play with refined team plans. ESWA, 298, 129496. (www)

  @article{ , abstract = {Adversarial team games represent scenarios where cooperation and competition coexist and hold numerous applications in the real world. These scenarios are particularly challenging due to asymmetric information among team members and limited communication capabilities. Fictitious team-play extend self-play algorithms to these scenarios, offering a novel approach to obtain equilibrium. However, it depends on normal-form team plans, which expand exponentially with game size, significantly constraining their applicability in large games. To overcome the challenge of computing equilibrium in large scale imperfect information team games, we propose a team self-play algorithm that utilizes refined team plans. Specifically, we pre-solve the equilibrium in a perfect recall environment to extract essential team plans from the original strategy space. To adapt these plans to an imperfect recall environment, we construct an auxiliary game with transformed ex ante coordinated information based on the original game and then solve equilibrium in auxiliary game to derive equilibrium for the original game. The experiments demonstrate the effectiveness of our team self-play algorithm in eight different Kuhn poker scenarios. Compared to existing team self-play algorithms, our method efficiently handles large games and exhibits superior convergence compared to reinforcement learning based algorithms. Additionally, our experiments offer valuable insights and guidance on adapting equilibrium strategies from perfect recall environments to those with imperfect recall.}, archive = {J_ESWA}, author = {Jinheng Xiao and Chen Qiu and Yingying Xu and Jiajia Zhang and Shuhan Qi and Xuan Wang}, doi = {10.1016/j.eswa.2025.129496}, journal = {Expert Systems with Applications}, month = {3}, pages = {129496}, shortjournal = {Expert Syst. Appl.}, title = {Solving equilibrium for adversarial team games utilizing fictitious team play with refined team plans}, volume = {298}, year = {2026}, }
• (2026). Proactive-BiLSTM: Glaucoma detection using FUNDUS and OCT retinal images. ESWA, 298, 129490. (www)

  @article{ , abstract = {Glaucoma detection identifies the early signs of eye conditions that can lead to vision loss by analyzing the retinal images to detect abnormalities, such as increased intraocular pressure, changes in the optic nerve head, or structural alterations in the retina. The challenges faced by the existing models include the difficulty in detecting subtle features, variability in image quality, and complex patterns that may resemble normal variations. Moreover, the traditional models struggle to adapt to the evolving patient data, capture long-term dependencies, and often suffer from lower accuracy. Hence, this research proposes the Proactive Hybridized Bidirectional Long Short-Term Memory (BiLSTM) model for Glaucoma detection. The proactive hybridized BiLSTM model is designed to enhance the detection of glaucoma by processing the retinal images. The proactive hybridized BiLSTM model enables the model to capture complex temporal dependencies and relationships within the data, which are crucial for identifying subtle patterns indicative of glaucoma, for which multifaceted feature extraction is employed. Moreover, the Proactive Hybridized BiLSTM model adapts to dynamic changes in the data to learn and predict glaucoma-related features, ultimately improving detection performance over time. The proposed Proactive hybridized BiLSTM model attains higher accuracy, sensitivity, and specificity of 96.65%, 96.51%, and 96.79% using the OCT and FUNDUS image dataset.}, archive = {J_ESWA}, author = {M.Kiran Mayee and M.Humera Khanam and Shaik Lathifa Tabasum}, doi = {10.1016/j.eswa.2025.129490}, journal = {Expert Systems with Applications}, month = {3}, pages = {129490}, shortjournal = {Expert Syst. Appl.}, title = {Proactive-BiLSTM: Glaucoma detection using FUNDUS and OCT retinal images}, volume = {298}, year = {2026}, }
• (2026). DBCD: Deep balanced community detection via consensus-guided joint optimization in attributed networks. ESWA, 298, 129487. (www)

  @article{ , abstract = {Current deep learning methods for community detection in attributed networks face a critical limitation: they often fail to identify communities that are both structurally cohesive and semantically similar, thereby falling short of the balance typically observed in human-labeled partitions. This shortcoming stems from the absence of explicit mechanisms to jointly optimize these two objectives. In this paper, this challenge is addressed by proposing Deep Balanced Community Detection (DBCD), a novel unsupervised framework for community detection that balances topology and semantics. DBCD first constructs a powerful topology-semantic clustering consensus by integrating insights from both structural and attribute spaces. This consensus then steers a Graph Neural Network to simultaneously maximize global neural modularity and local cross-view consistency, while adaptively determining the number of communities. Extensive experiments reveal a striking result: DBCD consistently discovers communities that surpass the topology-semantic balance of the ground truth across multiple real-world networks. An empirical Pareto frontier analysis further validates that DBCD achieves a non-dominated solution, establishing it as a strong competitor among state-of-the-art methods. The source code of DBCD is available at https://github.com/wy980125/DBCD .}, archive = {J_ESWA}, author = {Yan Wang and Yupeng Liu and Xiaojie Sun and Jun Fu}, doi = {10.1016/j.eswa.2025.129487}, journal = {Expert Systems with Applications}, month = {3}, pages = {129487}, shortjournal = {Expert Syst. Appl.}, title = {DBCD: Deep balanced community detection via consensus-guided joint optimization in attributed networks}, volume = {298}, year = {2026}, }
• (2026). Network traffic forecasting with transfer learning-based algorithm for long continuous missing data. ESWA, 298, 129484. (www)

  @article{ , abstract = {Accurate network traffic forecasting is critical for power dispatch networks. Network traffic forecasting aims to use historical data to predict future network traffic trends. Different from other networks, the traffic data of the power dispatch network is mainly composed of the port traffic from routers and switches. However, network accidents in power enterprises can cause long periods of missing network traffic data, reducing the number of learning samples for network traffic prediction models and making the forecasting results unreliable. Due to the long periods of missing data, this paper uses transfer learning (TL) to impute missing data with the knowledge from a relevant task, which has ample samples. However, the imputation result contains complex source and target data characteristics. Therefore, this paper introduces the idea of frequency decomposition to decompose the imputation results into different sub-sequences through variational mode decomposition (VMD). Additionally, this paper uses long short-term memory (LSTM) networks to extract the potential features of decomposition results. Finally, this paper combines TL, VMD, and LSTM to design the TL-VMD-LSTM algorithm. The effectiveness of the proposed algorithm is validated using inflow and outflow traffic data from two State Grid Corp. of China networks. The results demonstrate that TL-VMD-LSTM has excellent generalization performance, with mean absolute percentage errors (MAPEs) of 0.380 % and 0.734 % for the Provincial access network and Information region network, respectively.}, archive = {J_ESWA}, author = {Yang Yang and Zhihao Chen and Yuchao Gao and Zijin Wang and Zhe Ding and Jinran Wu}, doi = {10.1016/j.eswa.2025.129484}, journal = {Expert Systems with Applications}, month = {3}, pages = {129484}, shortjournal = {Expert Syst. Appl.}, title = {Network traffic forecasting with transfer learning-based algorithm for long continuous missing data}, volume = {298}, year = {2026}, }
• (2026). High-dimensional multi-objective feature selection with niche-based binary differential evolution. ESWA, 298, 129478. (www)

  @article{ , abstract = {Feature selection is a critical step in machine learning and data mining, aiming to identify the most relevant features from a dataset to improve model performance while reducing computational costs. In high-dimensional data, as the dimensionality of data increases rapidly, feature selection faces an enormous search space, limiting the efficiency and effectiveness of traditional methods. To address these challenges, multi-objective optimization algorithms have emerged as a promising strategy for feature selection due to their ability to optimize multiple conflicting objectives simultaneously. We propose a niche-based binary differential evolution algorithm (MONBDE) for high-dimensional multi-objective feature selection. MONBDE enhances feature selection performance through several mechanisms: a niche-based binary differential evolution operator, redundant solution repair mechanism and an environmental selection strategy. In experiments, the proposed algorithm was compared with five advanced multi-objective optimization algorithms and tested on 15 benchmark datasets using three common metrics. Experimental results show that the MONBDE algorithm outperforms comparative algorithms in terms of classification accuracy and feature subset size across most datasets. The proposed strategy effectively eliminates redundant and irrelevant solutions in feature selection, leading to a significant improvement in model classification performance.}, archive = {J_ESWA}, author = {Xuezhi Yue and Xiang Zuo and Pengfei Ling and Chao Xiong and Hu Peng and Yuan Zeng}, doi = {10.1016/j.eswa.2025.129478}, journal = {Expert Systems with Applications}, month = {3}, pages = {129478}, shortjournal = {Expert Syst. Appl.}, title = {High-dimensional multi-objective feature selection with niche-based binary differential evolution}, volume = {298}, year = {2026}, }
• (2026). DFCNet: Dual-path fusion with intra-slice features and cross-slice constraints for cervical cancer CTV segmentation. ESWA, 298, 129477. (www)

  @article{ , abstract = {Accurate and rapid segmentation of the clinical target volume (CTV) is essential for cervical cancer radiotherapy. However, due to the soft boundaries of CTV, complex connections with surrounding tissues, and high interpatient variability, existing deep learning methods still face significant challenges, particularly for image slices that lack clear boundary information or where applicators are distant from CTV edges. Hence, we introduce DFCNet, a dual-path fusion network with cross-slice consistency constraints for CTV segmentation. The first path employs a dual-stream intra-slice feature encoding module to capture local and inter-regional details, thereby refining boundary delineation amidst the complex interplay with adjacent tissues. The second path integrates a cross-slice consistency constraint module to address soft boundaries and high interpatient variability, while ensuring the coherence and smoothness of the segmentation results. A feature fusion and decoding module combines semantic features from both paths, improving CTV region accuracy. Tests on 432 cervical cancer brachytherapy cases show DFCNet outperforms eighteen state-of-the-art segmentation methods, with Dice score improvements over two percentage points. The second path and feature fusion module can enhance other U-Net-based models, boosting their CTV segmentation performance. DFCNet excels in high-precision CTV segmentation, particularly for challenging slices, demonstrating its potential to improve cervical cancer radiotherapy accuracy, efficiency, and patient outcomes.}, archive = {J_ESWA}, author = {Mingxu Huang and Deyu Sun and Chaolu Feng and Ming Cui and Dazhe Zhao and Yuhua Gao}, doi = {10.1016/j.eswa.2025.129477}, journal = {Expert Systems with Applications}, month = {3}, pages = {129477}, shortjournal = {Expert Syst. Appl.}, title = {DFCNet: Dual-path fusion with intra-slice features and cross-slice constraints for cervical cancer CTV segmentation}, volume = {298}, year = {2026}, }
• (2026). Competitive e-commerce platforms’ data provision and pricing strategies with different attribution behaviors of users. ESWA, 298, 129466. (www)

  @article{ , abstract = {In the era of data, numerous e-commerce platforms are actively collecting consumer data to obtain business insights. However, consumers and merchants on platforms exhibit diverse attribution behaviors, including single-homing and multi-homing (access to only one platform/multiple platforms), not only affecting the platform’s market scale but also complicating their data provision strategies and data pricing strategies. Inspired by this practice, this paper considers varying attribution behaviors and studies the data operation strategies of competitive platforms. By constructing a two-period game model, we capture the entire process of platform’s data collection and provision, and solve the equilibrium decisions by reverse solution method. This research aims to identify the impact of attribution behaviors on platforms’ data strategies, thereby filling the gap in analyzing this issue from the perspective of two-sided platforms. Results show that when both groups of users (consumers and merchants) are multi-homing, platform facing higher operational costs may benefit more from implementing low data provision but high data pricing strategy, while more cost-efficient platform may choose the opposite strategy. This strategy is still applicable when only one group of users (consumers) becomes single-homing. However, once both groups of users are single-homing, both platforms’ strategies will change. Specifically, when merchant’s cross-side network effect (CNE) intensity is relatively low (high), compared with less cost-efficient platform, platform enjoying cost efficiencies should provide more (less) data at a relatively high (low) price. Moreover, platforms should cautiously provide data when both groups of users are single-homing, as it may hurt profits.}, archive = {J_ESWA}, author = {Wei Chen and Yijia Hu and Ronghua Sui and Zili Guan and Yi Liu}, doi = {10.1016/j.eswa.2025.129466}, journal = {Expert Systems with Applications}, month = {3}, pages = {129466}, shortjournal = {Expert Syst. Appl.}, title = {Competitive e-commerce platforms’ data provision and pricing strategies with different attribution behaviors of users}, volume = {298}, year = {2026}, }
• (2026). Emergency logistics planning through optimizing the multi-trip vehicle routing with time windows and limited trip duration. ESWA, 298, 129465. (www)

  @article{ , abstract = {Although the multi-trip vehicle routing problem with time windows (MTVRPTW) and its variants have been extensively studied, their application in natural disaster contexts remains underexplored. This study addresses this gap by developing a model and solution algorithm for the MTVRPTW with limited trip duration (MTVRPTW-LD), tailored to emergency supplies distribution in the early post-disaster phase. First, we replace the service-dependent loading time in traditional models with service-dependent unloading time and formulate an MTVRPTW-LD model to minimize total operational time, encompassing travel, service, and unloading times, based on the characteristics of emergency supplies distribution. Furthermore, a more practical method for calculating travel time is proposed to enhance the model’s applicability. Subsequently, a branch-and-price algorithm is designed to solve the MTVRPTW-LD model, in which the cumulative relative deprivation cost (CRDC) is introduced to improve equity in emergency supplies distribution. Finally, we conduct numerical experiments on Solomon instances and test instances generated based on emergency scenarios. The results show that, in the test instances, incorporating CRDC can improve the equity by up to 34.3 %.}, archive = {J_ESWA}, author = {Longfei Fan and Zhongming Wu and Zaiwu Gong and David Z.W. Wang}, doi = {10.1016/j.eswa.2025.129465}, journal = {Expert Systems with Applications}, month = {3}, pages = {129465}, shortjournal = {Expert Syst. Appl.}, title = {Emergency logistics planning through optimizing the multi-trip vehicle routing with time windows and limited trip duration}, volume = {298}, year = {2026}, }
• (2026). Regularity model-driven large-scale multi-objective evolutionary algorithm based on dual-information offspring reproduction strategy. ESWA, 298, 129460. (www)

  @article{ , abstract = {The Pareto set (PS) of a continuous multi-objective optimization problem exhibit a distribution along a low-dimensional manifold structure. This regularity property significantly contributes to generating high-quality offspring in large-scale multi-objective evolutionary algorithms (LSMOEAs). However, conventional regularity model-based algorithms face several challenges when dealing with large-scale multi-objective optimization problems (LSMOPs), including high computational costs for modeling, difficulty in capturing the true PS structure, and neglecting individual directional information. To address these challenges, we propose a dual-information offspring reproduction strategy that considers both the distribution information of the population and the directional information of the outstanding individuals. Specifically, this strategy comprises a sampling approach based on an augmented regularity model specifically designed for LSMOPs. Leveraging this model, we explore and exploit the decision space to sample a promising set of solutions. Additionally, the strategy also involves a search method based on competitive learning among individuals. By assigning a positive evolutionary direction to losing solutions, we update the losing solutions to generate high-quality offspring. We continuously refine the proposed regularity model to approximate the true PS more closely. In extensive experiments on large-scale multi-objective benchmark functions, we compare our algorithm with eight state-of-the-art algorithms. The results demonstrate that our approach excels in handling LSMOPs.}, archive = {J_ESWA}, author = {Ying Wu and Ziliang Du and Gonglin Yuan and Zhenzhou Tang and Ferrante Neri and Yaqing Hou}, doi = {10.1016/j.eswa.2025.129460}, journal = {Expert Systems with Applications}, month = {3}, pages = {129460}, shortjournal = {Expert Syst. Appl.}, title = {Regularity model-driven large-scale multi-objective evolutionary algorithm based on dual-information offspring reproduction strategy}, volume = {298}, year = {2026}, }
• (2026). GlobalCLIP: Zero-shot manufacturing anomaly detection with adaptive self-cyclic emsemble learning. ESWA, 298, 129448. (www)

  @article{ , abstract = {Accurately identifying product defects and process anomalies in manufacturing processes is a critical task for product quality and system stability. Although the existing unsupervised anomaly detection methods do not require the annotation of anomalies, they are difficult to deal with the zero-shot scenario faced by multi-variety and small-batch production where there is no available data. In addition, many zero-shot detection algorithms need to represent the image features in tensor form with multiple local feature vectors, and then measure each local feature to infer the overall anomaly of the object. In this paper, we propose GlobalCLIP, a novel approach for zero-shot anomaly detection using only global feature vectors to enhance performance. Specifically, we use CLIP model to aggregate the global features, and design two kinds of adaptive modules from the error level and uncertainty level to realize the series integration of different discriminant models. The adaptive modules encourage the model to learn both normal and abnormal patterns with different granularity, and the self-cyclic training progressively improves model performance. Experiments show that compared to many unsupervised/weakly supervised methods, the performance of GlobalCLIP maintains its advantage even without known samples, and achieves significant improvement over available zero-shot methods.}, archive = {J_ESWA}, author = {Haoyuan Shen and Enrico Zio and Jiawei Xiong and Yizhong Ma}, doi = {10.1016/j.eswa.2025.129448}, journal = {Expert Systems with Applications}, month = {3}, pages = {129448}, shortjournal = {Expert Syst. Appl.}, title = {GlobalCLIP: Zero-shot manufacturing anomaly detection with adaptive self-cyclic emsemble learning}, volume = {298}, year = {2026}, }
• (2026). Key performance indicator-related process monitoring for irregular scenarios with incomplete data. ESWA, 298, 129440. (www)

  @article{ , abstract = {Modern industries exhibit irregular characteristics due to factors such as mode transitions, incomplete data and outliers. Accurate monitoring of key performance indicators (KPIs) in irregular processes is essential for improving product quality and reducing scrap rates. This paper proposes a novel KPI-related process monitoring method that leverages the multiple kernel learning (MKL) technique, designed specifically for irregular scenarios with incomplete data. First, a novel MKL-based nonlinear matrix completion is proposed that utilizes a hierarchical strategy-based algorithm to estimate the missing values in incomplete data and the linear coefficients of multiple kernels. In addition, the corresponding convergence analysis is given. Based on the estimated completed data matrix, a novel MKL-based feature correlation analysis is proposed for indirect prediction of KPIs. Two statistics are established for detecting KPI-related and KPI-unrelated faults, respectively. A numerical case and an industrial example demonstrate that the proposed method not only accurately identifies the missing data, but also effectively detects the KPI-related faults.}, archive = {J_ESWA}, author = {Yanyu Chen and Hao Ma and Yan Wang and Xiang Liu}, doi = {10.1016/j.eswa.2025.129440}, journal = {Expert Systems with Applications}, month = {3}, pages = {129440}, shortjournal = {Expert Syst. Appl.}, title = {Key performance indicator-related process monitoring for irregular scenarios with incomplete data}, volume = {298}, year = {2026}, }
• (2026). A novel approach to anomaly detection in critical CPSs: The ERXAI-MS algorithm. ESWA, 298, 129437. (www)

  @article{ , abstract = {At present, practical machine learning methodologies are extensively employed for the automation of detecting the intrusion available in the network. In key infrastructure scenarios involving communication strategies, the interplay between different industrial control systems and the inherent connection to the Internet environment through the Internet of Things renders them vulnerable to cyber threats. Considering the substantial volume of network traffic within critical Cyber-Physical Systems, conventional machine-learning approaches utilized for detecting anomalies prove to be ineffective. Hence, newly designed machine learning methods, with a focus on deep learning, are demonstrating effective applications in identifying and categorizing anomalies on both network and individual device levels. This article introduces an innovative Ensemble Random Explainable Artificial Intelligence Meerkat Search algorithm designed for the identification of cyber threats. To augment the effectiveness of the suggested method, it employs a dual-step process for the detection of network irregularities. During the initial phase, the approach involves data pre-processing and dimensionality reduction through the application of Kernel Principal Component Analysis to select the most suitable features. In the subsequent stage, the novel Ensemble Random Explainable Artificial Intelligence Meerkat Search algorithm is employed for classification. The effectiveness of the approach presented in this study is evaluated on diverse datasets, encompassing information collected within the Internet of Things context, specifically IoT-23 and LITNET-2020 datasets. The findings of the assessment of the suggested method are deliberated upon, including the examination of statistical significance and a comparative analysis with contemporary approaches in the field of network anomaly detection. Evaluations confirmed this robust model attained 98.56% accuracy, 97.78% precision, 98.2% F1-score, and produced less FPR of 1.55%.}, archive = {J_ESWA}, author = {Prabakeran Saravanan and Annamalai Balaji and Hemalatha Murugan and Manickam Muruganantham and Indumathi Varadharajan}, doi = {10.1016/j.eswa.2025.129437}, journal = {Expert Systems with Applications}, month = {3}, pages = {129437}, shortjournal = {Expert Syst. Appl.}, title = {A novel approach to anomaly detection in critical CPSs: The ERXAI-MS algorithm}, volume = {298}, year = {2026}, }
• (2026). Improved DQN-based recommender system on three-way decision. ESWA, 298, 129431. (www)

  @article{ , abstract = {Recommendation systems, which utilize algorithms and data analysis to provide personalized suggestions to users, have become an indispensable part of modern life. However, traditional recommendation algorithms face challenges such as the cold start problem, lack of diversity, and limited scalability. Reinforcement learning (RL), particularly deep reinforcement learning (DRL), emerges as a promising solution to these problems by allowing agents to learn optimal strategies through interaction with their environment. Nevertheless, as the scale of data increases, RL-based recommendation systems often struggle to achieve a good balance between exploration and exploitation, impacting the overall performance of the algorithms. In this paper, we propose a reinforcement learning-based recommendation algorithm enhanced by a three-way decision (3WD) framework to address the exploration-exploitation balance challenge. The 3WD algorithm, rooted in rough set theory, categorizes decision outcomes into acceptance, rejection, and uncertainty regions. By applying 3WD in the action selection process of RL, we optimize the trade-off between exploration and exploitation, thereby improving the quality and computational efficiency of recommendations. Additionally, we introduce a dynamic threshold adjustment mechanism to adaptively refine the decision boundary during the action selection process in reinforcement learning, further enhancing the algorithm’s performance. Using the MovieLens dataset as a foundation, we conduct extensive experiments with several randomly generated data sets to evaluate the proposed method. Our results demonstrate that the 3WD-based RL algorithm outperforms traditional methods, such as epsilon-greedy and Softmax, in terms of runtime, recommendation accuracy, and error rate. Notably, the dynamic threshold adjustment model exhibits greater stability and surpasses static methods in recommendation success rates. These findings highlight the effectiveness of combining 3WD with RL in recommendation systems, providing a powerful and efficient solution to the challenges faced by traditional methods. Finally, we analyze the limitations of the model based on the experimental results and propose avenues for future research.}, archive = {J_ESWA}, author = {Zian Chen and Bao Qing Hu}, doi = {10.1016/j.eswa.2025.129431}, journal = {Expert Systems with Applications}, month = {3}, pages = {129431}, shortjournal = {Expert Syst. Appl.}, title = {Improved DQN-based recommender system on three-way decision}, volume = {298}, year = {2026}, }
• (2026). IntegralGP: Volumetric estimation of subterranean geochemical properties in mineral deposits by fusing assay data with different spatial supports. ESWA, 298, 129429. (www)

  @article{ , abstract = {This article presents an Integral Gaussian Process (IntegralGP) framework for volumetric estimation of subterranean properties in mineral deposits. It provides a unified representation for data with different spatial supports, which enables blasthole geochemical assays to be properly modelled as interval observations rather than points. This approach is shown to improve regression performance and boundary delineation. A core contribution is a description of the mathematical changes to the covariance expressions which allow these benefits to be realised. The gradient and anti-derivatives are obtained to facilitate learning of the kernel hyperparameters. Numerical stability issues are also discussed. To illustrate its application, an IntegralGP data fusion algorithm is described. The objective is to assimilate line-based blasthole assays and update a block model that provides long-range prediction of Fe concentration beneath the drilled bench. Heteroscedastic GP is used to fuse chemically compatible but spatially incongruous data with different resolutions and sample spacings. Domain knowledge embodied in the structure and empirical distribution of the block model must be generally preserved while local inaccuracies are corrected. Using validation measurements within the predicted bench, our experiments demonstrate an improvement in bench-below grade prediction performance. For material classification, IntegralGP fusion reduces the absolute error and model bias in categorical prediction, especially instances where waste blocks are mistakenly classified as high-grade.}, archive = {J_ESWA}, author = {Anna Chlingaryan and Arman Melkumyan and Raymond Leung}, doi = {10.1016/j.eswa.2025.129429}, journal = {Expert Systems with Applications}, month = {3}, pages = {129429}, shortjournal = {Expert Syst. Appl.}, title = {IntegralGP: Volumetric estimation of subterranean geochemical properties in mineral deposits by fusing assay data with different spatial supports}, volume = {298}, year = {2026}, }
• (2026). Deep multi-view least squares support vector machine with consistency and complementarity principle based on cross-output knowledge transfer. ESWA, 298, 129406. (www)

  @article{ , abstract = {This article presents a Deep Multi-view Least Squares Support Vector Machine with Consistency and Complementarity Principles based on Cross-Output Knowledge Transfer (MDCTM), which has four distinctive features: 1) It integrates the idea of deep stacking architecture, which is the first attempt to use transfer learning to form deep architectures in multi-view learning. It can enhance the ability to handle complex problems. Starting from the second layer, it incorporates extra input attributes that consider the predictions made by all preceding layers, effectively revealing the manifold structure of the original data. 2) Each layer follows the consistency and complementarity principles, which can fully excavate the information in multi-view data. In each layer, the model is solved by an alternating optimization strategy. 3) Cross-output knowledge transfer leverages predictions from earlier layers to improve the learning of subsequent ones, which can improve the classification performance of the model. Additionally, the extent of cross-output knowledge transfer between sequential layers can be assessed autonomously and effectively by utilizing a fast leave-one-out cross-validation method. 4) The model allows random assignment of model parameters in each layer, such as weights and kernel widths, boosting learning speed. Numerical experiments demonstrate the model’s effectiveness and efficiency.}, archive = {J_ESWA}, author = {Shuangrui Jia and Sijie Liang and Ziyi Mo and Chunxiao Liu and Huiru Wang and Chen Chen}, doi = {10.1016/j.eswa.2025.129406}, journal = {Expert Systems with Applications}, month = {3}, pages = {129406}, shortjournal = {Expert Syst. Appl.}, title = {Deep multi-view least squares support vector machine with consistency and complementarity principle based on cross-output knowledge transfer}, volume = {298}, year = {2026}, }
• (2026). A study on the modeling of long-term trend and short-term fluctuation: Fourier-enhanced adaptive koopman operator for carbon emission forecasting. ESWA, 298, 129388. (www)

  @article{ , abstract = {The Koopman operator provides a new way to model complex data patterns, revealing the intrinsic dynamics of time series from a dynamical system perspective. Despite its potential, the Koopman operator has received limited attention in time series forecasting, particularly in addressing these complex, real-world challenges such as carbon emission dynamics characterized by nonlinearity, non-stationarity, and multi-scale coupling effects. To this end, this study proposes a novel forecasting paradigm, Fourier-Enhanced adaptive Koopman operator for carbon emission forecasting (F-KOCE). This approach conceptually extends traditional Koopman frameworks by embedding a spectral-decoupled time series representation into a dual Koopman learning structure, which enables the model to linearize nonlinear dynamics across multiple time scales in a theoretically grounded and practically adaptive manner. By integrating Fourier filter decomposition into Koopman operator theory, F-KOCE separates raw emissions into long-term trends and short-term fluctuations while achieving global linearization of system dynamics. A learnable Koopman operator captures intrinsic temporal structures, while a multi-granularity adaptive weight learning strategy enhances resilience against data variability. To further improve robustness, we introduce an adaptive residual fusion structure for block-level feature compression, noise suppression, and cross-scale information fusion. Additionally, the effective Trend Corrector mechanism dynamically modulates the influence of trend and fluctuation components, refining predictive accuracy. Beyond point forecasting, the framework is also extended to interval forecasting, providing uncertainty-aware predictions. Extensive experiments conducted on 36 Carbon Monitor datasets across six regions and six sectors demonstrate the superiority of F-KOCE over advanced existing models across multiple evaluation metrics. These results confirm the framework’s efficacy in capturing high-dimensional emission dynamics and underscore the potential of Koopman operator theory in carbon forecasting. By offering a robust, interpretable, and data-driven approach, F-KOCE provides valuable insights for climate policy formulation.}, archive = {J_ESWA}, author = {Jinxing Che and Wei Dong and Qian Sun and Yuhua Zhang}, doi = {10.1016/j.eswa.2025.129388}, journal = {Expert Systems with Applications}, month = {3}, pages = {129388}, shortjournal = {Expert Syst. Appl.}, title = {A study on the modeling of long-term trend and short-term fluctuation: Fourier-enhanced adaptive koopman operator for carbon emission forecasting}, volume = {298}, year = {2026}, }
• (2026). A cooperative co-evolutionary algorithm with core-based grouping strategy for large-scale 0–1 knapsack problems. ESWA, 298, 129364. (www)

  @article{ , abstract = {The 0–1 knapsack problem (KP) is a well-known combinatorial optimization problem with wide real-world applications. While evolutionary algorithms have demonstrated promise in solving 0–1 KPs, their performance deteriorates as the problem dimension increases. Cooperative co-evolution (CC) is an algorithmic framework based on a divide-and-conquer strategy, which has been used in solving large-scale optimization problems. Inspired by the similarity between item grouping in the 0–1 KP and decomposition strategies in CC, this paper proposes a novel grouping strategy that uses the position information of break items and profit-to-weight ratio to solve large-scale 0–1 KP. The strategy aims to divide the large-scale 0–1 KP into multiple subproblems, thus having a reduced search space for each subproblem. To enhance population diversity and search efficiency, the profit-to-weight ratio is used to generate an initial elite population. Additionally, to obtain the complete solution for the original large-scale KP, a subgroup merging method is designed to accelerate convergence and further improve population diversity. A three-phase repair operator is developed to fix infeasible solutions directly to create more feasible solutions. The resulting cooperative co-evolutionary algorithm is compared with ten state-of-the-art algorithms for solving 0–1 KPs with variables ranging from 100 to 5,000, including EAs, CC-based approaches, and a deep reinforcement learning method. Experimental results show that the proposed algorithm exhibits higher solution accuracy and faster convergence than other competing algorithms. The CC framework takes considerably less running time than high-performing algorithms, providing an overall novel approach for solving large-scale 0–1 KPs.}, archive = {J_ESWA}, author = {Xiaotong Li and Shuwei Zhu and Wei Fang and Kalyanmoy Deb}, doi = {10.1016/j.eswa.2025.129364}, journal = {Expert Systems with Applications}, month = {3}, pages = {129364}, shortjournal = {Expert Syst. Appl.}, title = {A cooperative co-evolutionary algorithm with core-based grouping strategy for large-scale 0–1 knapsack problems}, volume = {298}, year = {2026}, }
• (2026). Balancing forecast accuracy and switching costs in online optimization of energy management systems. ESWA, 298, 129305. (www)

  @article{ , abstract = {This study investigates the integration of forecasting and optimization in energy management systems, focusing on how switching costs, penalties incurred from frequent operational adjustments, affect the balance between forecast accuracy and stability in online decision-making. We develop a theoretical framework analyzing Fixed Horizon Control (FHC) algorithms under switching costs, deriving performance bounds that reveal trade-offs between commitment periods and forecast properties. We introduce a novel Scenario Distribution Change (SDC) metric for measuring temporal consistency in probabilistic forecasts. The framework is validated through empirical evaluation using a real-world battery scheduling case study based on the CityLearn 2022 challenge, comparing deterministic and stochastic optimization approaches across different commitment periods. Theoretical analysis reveals that switching costs create a U-shaped relationship between commitment period and performance, with optimal commitment depending on forecast stability. Empirical results demonstrate that switching costs significantly alter the accuracy-stability trade-off: while traditional approaches favor frequent updates (1-hour commitment), incorporating switching costs makes longer commitment periods (3+ hours) optimal when combined with stable forecasts. Stochastic optimization with scenario averaging reduces forecast error sensitivity by up to 2.9 % in grid costs compared to deterministic approaches. This work contributes the first theoretical bounds linking forecast stability to switching costs in energy systems, the SDC metric for evaluating probabilistic forecast stability, empirical evidence that longer commitment periods can outperform frequent updates under switching costs, and practical guidelines showing that forecast stability should be factored into decision-making frameworks for energy management systems in the presence of switching costs.}, archive = {J_ESWA}, author = {Evgenii Genov and Julian Ruddick and Christoph Bergmeir and Majid Vafaeipour and Thierry Coosemans and Salvador García and Maarten Messagie}, doi = {10.1016/j.eswa.2025.129305}, journal = {Expert Systems with Applications}, month = {3}, pages = {129305}, shortjournal = {Expert Syst. Appl.}, title = {Balancing forecast accuracy and switching costs in online optimization of energy management systems}, volume = {298}, year = {2026}, }
• (2026). A hybrid bee colony algorithm for crack repair trajectory planning based on focal attention guided lightweight segmentation. ESWA, 298, 129267. (www)

  @article{ , abstract = {Robotic arm trajectory planning of crack repair is critical for automated road maintenance. However, existing crack repair face two main challenges: loss of trajectory edge information and redundant planned distances. This study introduces an automated pavement crack repair system that integrates a lightweight crack segmentation model (Lightweight Focal Modulation, LFM-Net) and a repair trajectory planning algorithm (Fixed Neighborhood Search-Artificial Bee Colony, FNS-ABC). Specifically, LFM-Net incorporates conformer-based focal modulation attention (CFMA), enhancing the detailed information during the decoding phase. Additionally, the FNS-ABC enhances the ABC algorithm by incorporating a fixed neighborhood search strategy, effectively reducing redundant planning paths. The system is executed using a self-developed robotic arm with an edge computing unit. Extensive testing in three typical road scenarios-independent cracks, intersection cracks, and complex cracks-demonstrated that the system achieved a mean Intersection over Union (mIoU) of 83.93 %. Finally, the system exhibited an idle trajectory of 79.51 mm when addressing complex cracks, highlighting its superior performance in repair trajectory planning.}, archive = {J_ESWA}, author = {Jianqi Zhang and Xu Yang and Wei Wang and Yuhang Zhao and Hainian Wang and Yixue Chen}, doi = {10.1016/j.eswa.2025.129267}, journal = {Expert Systems with Applications}, month = {3}, pages = {129267}, shortjournal = {Expert Syst. Appl.}, title = {A hybrid bee colony algorithm for crack repair trajectory planning based on focal attention guided lightweight segmentation}, volume = {298}, year = {2026}, }
• (2026). Deep-learning based big data analysis for developing a smart supply chain for increased efficiency. ESWA, 298, 129246. (www)

  @article{ , abstract = {Big data analysis (BDA) in supply chain management (SCM) is receiving growing attention in the present business environment. This is due to the fact that BDA has a wide range of applications in SCM, including customer behaviour analysis, trend analysis, and demand prediction. The increase in information volume has caused the efficiency and effectiveness of traditional procedures to decline, considering this, researchers have developed techniques that have a high capacity to investigate and comprehend vast amounts of data due to the limitations of these tactics in dissecting and interpreting a lot of information. This study represents a hybrid paper that combines a systematic literature review, a methodological proposal using BP neural networks. The main objective of this paper is to recognize the uses of deep learning in SCM. By fostering a calculated system, this paper recognizes the commitments of deep learning strategies in choosing and sectioning providers, foreseeing store network gambles, and assessing requests and deals, creation, stock administration, transportation and circulation, manageable turn of events, and roundabout economy. The novelty in this paper is the Backpropagation (BP) neural networks with big data-driven demand forecasting in supply chains. This method can improve the accuracy of demand forecasting in supply chain management. The study includes a thorough survey of the applications of predictive BDA in SC request gauging. The review highlighted the BDA methodologies used for production network request estimation and comparatively classified them. We collected and analysed these studies as tactics and methodologies for the popular forecast. Seven standard tactics were selected and studied, along with their benefits and drawbacks. Finally, the box-cox transformation representation over years in which for the year 2011–01, it starts with a box-cox value of 9.7 and it inclined till 2011–06 and then declined very exponentially in 2012–07 at 9 and then it keeps on incrementing and reached at 11 at the year 2013–07. Then from 2014 to 2015, the pattern didn’t lower below the box-cox value 10.}, archive = {J_ESWA}, author = {Sreekumar Narayanan and Sudhir Ramadass and K. Thilagavathi and Rajiv Kumar}, doi = {10.1016/j.eswa.2025.129246}, journal = {Expert Systems with Applications}, month = {3}, pages = {129246}, shortjournal = {Expert Syst. Appl.}, title = {Deep-learning based big data analysis for developing a smart supply chain for increased efficiency}, volume = {298}, year = {2026}, }

IANDC - 11

• (2025). K-universality of regular languages. IANDC, 307, 105357. (www)

  @article{ , abstract = {A subsequence of a word w is a word u such that u = w [ i 1 ] w [ i 2 ] … w [ i k ] , for some set of indices 1 ≤ i 1 < i 2 < … < i k ≤ | w | . A word w is k -subsequence universal over an alphabet Σ if every word in Σ k appears in w as a subsequence. In this paper, we study the intersection between the set of k -subsequence universal words over some alphabet Σ and regular languages over Σ. We call a regular language L k- ∃ -subsequence universal if there exists a k -subsequence universal word in L , and k- ∀ -subsequence universal if every word of L is k -subsequence universal. We give algorithms solving the problems of deciding if a given regular language, represented by a finite automaton recognising it, is k- ∃ -subsequence universal and, respectively, if it is k- ∀ -subsequence universal , for a given k . The algorithms are FPT w.r.t. the size of the input alphabet, and their run-time does not depend on k ; they run in polynomial time in the number n of states of the input automaton when the size of the input alphabet is O ( log ⁡ n ) . Moreover, we show that the problem of deciding if a given regular language is k- ∃ -subsequence universal is NP-complete, when the language is over a large alphabet. Further, we provide algorithms for counting the number of k -subsequence universal words (paths) accepted by a given deterministic (respectively, non-deterministic) finite automaton, and ranking an input word (path) within the set of k -subsequence universal words accepted by a given finite automaton.}, archive = {J_IANDC}, author = {Duncan Adamson and Pamela Fleischmann and Annika Huch and Tore Koß and Florin Manea and Dirk Nowotka}, doi = {10.1016/j.ic.2025.105357}, journal = {Information and Computation}, month = {11}, pages = {105357}, shortjournal = {Inf. Comput.}, title = {K-universality of regular languages}, volume = {307}, year = {2025}, }
• (2025). Recent algorithmic advances in simple temporal networks with uncertainty: From faster controllability checking to faster execution. IANDC, 307, 105356. (www)

  @article{ , abstract = {This paper advances the state of the art in the dynamic controllability (DC) and dispatchability of Simple Temporal Networks with Uncertainty (STNUs) through four key contributions. First, findSRNC is an algorithm that identifies semi-reducible negative cycles in non-dynamically controllable STNUs. Running in O ( m n + k 2 n + k n log ⁡ n ) time (matching the fastest DC-checking algorithms), it handles repeated edges and uses polynomial space, even when cycles might contain exponentially many edges. Second, minDisp ESTNU + is an algorithm that improves dispatchability computation for STNUs from O ( k n 3 ) to O ( n 3 + k 2 n log ⁡ n ) time. It outputs dispatchable Extended STNUs (ESTNUs) having minimal numbers of edges, which is crucial for subsequent real-time execution. Third, the Canonical Form of Nested Diamond Structures in Dispatchable ESTNUs is a rigorous theory that facilitates correctness proofs for dispatchability algorithms. It also helped reveal and correct a flaw in a previously published algorithm. Fourth, our empirical evaluation using improved open-source implementations demonstrates the practical effectiveness of our algorithms. These contributions address fundamental computational bottlenecks in temporal planning systems, enabling more efficient reasoning about uncertain timing constraints while providing real-time guarantees required for robotics, scheduling, and automated planning applications.}, archive = {J_IANDC}, author = {Luke Hunsberger and Roberto Posenato}, doi = {10.1016/j.ic.2025.105356}, journal = {Information and Computation}, month = {11}, pages = {105356}, shortjournal = {Inf. Comput.}, title = {Recent algorithmic advances in simple temporal networks with uncertainty: From faster controllability checking to faster execution}, volume = {307}, year = {2025}, }
• (2025). Polynomial turing compressions for some graph problems parameterized by modular-width. IANDC, 307, 105355. (www)

  @article{ , abstract = {A polynomial Turing compression (PTC) for a parameterized problem L is a polynomial time Turing machine that has access to an oracle for a problem L ′ such that a polynomial in the input parameter bounds each query. Meanwhile, a polynomial compression (PC) can be regarded as a restricted variant of PTC where the machine can query the oracle exactly once and must output the same answer as the oracle. Bodlaender et al. (ICALP 2008) and Fortnow and Santhanam (STOC 2008) initiated an impressive hardness theory for PC under the assumption coNP ⊈ NP/poly. Let C be the set of all problems with PTCs but without PCs assuming coNP ⊈ NP/poly. Fernau et al. (STACS 2009) identified Leaf Out-tree( k ) as the first problem in C . However, little is known about C , with only a dozen problems confirmed in it over the last fifteen years. Open questions remain, such as whether CNF-SAT( n ) and k -path are in C , requiring novel ideas to clarify the differences between PTCs and PCs. In this paper, we enrich our knowledge about C by demonstrating that 17 problems parameterized by modular-width ( mw ), such as Chromatic Number( mw ) and Hamiltonian Cycle( mw ) , belong to C . Additionally, we develop a general recipe to prove the existence of PTCs for a class of problems, including these 17.}, archive = {J_IANDC}, author = {Weidong Luo}, doi = {10.1016/j.ic.2025.105355}, journal = {Information and Computation}, month = {11}, pages = {105355}, shortjournal = {Inf. Comput.}, title = {Polynomial turing compressions for some graph problems parameterized by modular-width}, volume = {307}, year = {2025}, }
• (2025). On cardinalities of rogers semilattices for families in the ershov hierarchy. IANDC, 307, 105354. (www)

  @article{ , abstract = {The theory of numberings provides classification results for families of sets in various computability-theoretic hierarchies. The algorithmic content of numberings is typically calibrated via the reducibility between numberings. For a given family of sets S , this reducibility gives rise to an upper semilattice of degrees that is often called the Rogers semilattice of S . This paper studies the cardinalities of Rogers semilattices for families of sets at finite levels of the Ershov hierarchy. The classical result of Khutoretskii (1971) shows that the Rogers semilattice of a family of c.e. sets is either one-element or countably infinite. Badaev and Lempp (2009) constructed a family of d.c.e. sets that demonstrates that the methods of Khutoretskii cannot be applied to obtain a similar result for Rogers semilattices already at the second level of the Ershov hierarchy. We prove that for any finite family of sets S at any finite level of the Ershov hierarchy, the corresponding Rogers semilattice is either one-element or countably infinite. We also obtain another sufficient condition for a Rogers semilattice to be infinite. This condition implies that the Rogers semilattice of Badaev and Lempp is also infinite.}, archive = {J_IANDC}, author = {Keng Meng Ng and Nikolay Bazhenov and Birzhan Kalmurzayev and Dias Nurlanbek}, doi = {10.1016/j.ic.2025.105354}, journal = {Information and Computation}, month = {11}, pages = {105354}, shortjournal = {Inf. Comput.}, title = {On cardinalities of rogers semilattices for families in the ershov hierarchy}, volume = {307}, year = {2025}, }
• (2025). Exact counting of subtrees with diameter no more than d in trees: A generating function approach. IANDC, 307, 105353. (www)

  @article{ , abstract = {Network motifs, regarded as fundamental building blocks, offer crucial insights into the structure and function of complex networks, with broad applications across disciplines including sociology, computer science, bioinformatics, chemoinformatics, and pharmaceutics. However, the identification of network motifs remains a significant and computationally challenging problem. Among various motifs, subtree enumeration has garnered substantial attention in recent years, particularly due to its relevance in network science and bioinformatics. For an n -vertex tree T , by introducing novel generating functions with ( d + 2 ) variables, we propose an innovative algorithm for the exact enumeration of T 's subtrees rooted at fixed vertex v , where the distance between v and the farthest leaf is k = 0 , 1 , … , d , and the distance between any two leaves is no more than d . Building on this algorithm, we develop novel recursive algorithms for exact enumerating various diameter no more than d subtrees (abbreviated as DNMT- d subtrees) of T . As applications, we apply these algorithms to derive the number of DNMT- d subtrees in a full binary tree B h with h ≥ 2 levels, and briefly discuss the density of DNMT- d subtrees in general trees. Our research generalizes the work of Frank Ruskey on Listing and Counting Subtrees of a Tree in 1981 and makes it a special case of our study where d equals the diameter of the tree T . Moreover, the proposed O ( d n 2 ) algorithms introduce new approaches for enumerating subtrees under diameter constraints and lay the groundwork for counting diameter-constrained subgraphs (motifs) in complex networks.}, archive = {J_IANDC}, author = {Yu Yang and Bang-Bang Jin and Xiaoming Sun and Xiao-Dong Zhang and Bo Li and Kai Zhao and Hua Wang}, doi = {10.1016/j.ic.2025.105353}, journal = {Information and Computation}, month = {11}, pages = {105353}, shortjournal = {Inf. Comput.}, title = {Exact counting of subtrees with diameter no more than d in trees: A generating function approach}, volume = {307}, year = {2025}, }
• (2025). Decision problems for systems of language equations and inequations. IANDC, 307, 105344. (www)

  @article{ , abstract = {Systems of language equations φ ( X 1 , … , X n ) = ψ ( X 1 , … , X n ) and inequations φ ( X 1 , … , X n ) ≠ ψ ( X 1 , … , X n ) are studied, where φ and ψ may contain Boolean operations and concatenation. It is proved that the problem whether such a system has a solution is Σ 2 0 -complete in the arithmetical hierarchy (cf. the earlier studied case of equations only, where it is co-r.e.-complete), the problem whether it has a unique solution is in Σ 3 0 ∩ Π 3 0 , and is both Σ 2 0 -hard and Π 2 0 -hard, existence of a finite or regular solution is an r.e.-complete problem, while testing whether a system has finitely many solutions is Σ 3 0 -complete. Furthermore, it is shown that the class of languages representable by unique solutions of such systems is exactly the class of recursive sets, but decision procedures for the set cannot be algorithmically constructed out of a system. All results hold already for equations over a unary alphabet.}, archive = {J_IANDC}, author = {Alexander Okhotin}, doi = {10.1016/j.ic.2025.105344}, journal = {Information and Computation}, month = {11}, pages = {105344}, shortjournal = {Inf. Comput.}, title = {Decision problems for systems of language equations and inequations}, volume = {307}, year = {2025}, }
• (2025). Towards a theoretical understanding of why local search works for clustering with fair-center representation. IANDC, 307, 105343. (www)

  @article{ , abstract = {The representative k -median problem generalizes the classical clustering formulations in that it partitions the data points into ℓ disjoint demographic groups and imposes a lower-bound constraint on the number of opened facilities from each group, such that all the groups are fairly represented by the opened facilities. Due to its simplicity, the local-search heuristic, which iteratively swaps a bounded number of closed facilities for the same number of opened ones to improve the solution, has been frequently used in the representative k -median problem. It is known that the local-search heuristic, when restricted to constant-size swaps, yields a constant-factor approximation if ℓ = 2 , and has an unbounded approximation ratio if ℓ is super-constant. However, for any constant ℓ > 2 , the existence of a constant-factor approximation under constant-size swaps remained an open question for a long time. In response to this question, we demonstrate that the local-search heuristic guarantees a ( 4 ℓ + 5 ) -approximation when up to ℓ ( ℓ + 1 ) facilities are allowed to be swapped in each iteration, thus providing an affirmative answer to the question. Our main technical contribution is a novel approach for theoretically analyzing the local-search heuristic, which bounds its approximation ratio by linearly combining the clustering cost increases induced by a set of hierarchically organized swaps. Our techniques also generalize to the k -means clustering formulation and reveal similar approximation guarantees for the local-search heuristic.}, archive = {J_IANDC}, author = {Zhen Zhang and Junfeng Yang and Limei Liu and Xuesong Xu and Guozhen Rong and Qilong Feng}, doi = {10.1016/j.ic.2025.105343}, journal = {Information and Computation}, month = {11}, pages = {105343}, shortjournal = {Inf. Comput.}, title = {Towards a theoretical understanding of why local search works for clustering with fair-center representation}, volume = {307}, year = {2025}, }
• (2025). The billaud conjecture for alphabet size 4. IANDC, 307, 105342. (www)

  @article{ , abstract = {The Billaud Conjecture, first stated in 1993, is a fundamental problem on finite words and their heirs, i.e., the words obtained by a projection deleting a single letter. The conjecture states that every morphically primitive word, i.e., a word that is not a fixed point of any non-identity morphism, has at least one morphically primitive heir. The correctness of the conjecture has so far been established in a few special cases, which mainly restrict the alphabet size. In this paper we give a proof for the next such case, i.e., for alphabet size 4.}, archive = {J_IANDC}, author = {Szymon Łopaciuk and Daniel Reidenbach}, doi = {10.1016/j.ic.2025.105342}, journal = {Information and Computation}, month = {11}, pages = {105342}, shortjournal = {Inf. Comput.}, title = {The billaud conjecture for alphabet size 4}, volume = {307}, year = {2025}, }
• (2025). The g-good-neighbor diagnosability of product networks under the PMC model. IANDC, 307, 105341. (www)

  @article{ , abstract = {The concept of neighbor connectivity originated from the assessment of the subversion of espionage networks caused by underground resistance movements, and it has now been applied to measure the disruption of networks caused by cascading failures through neighbors. In this paper, we give two necessary and sufficient conditions of the existence of g -good-neighbor diagnosability. We introduce a new concept called g -good neighbor cut-component number (gc number for short), which has close relation with g -good-neighbor diagnosability. Sharp lower and upper bounds of the gc number of general graphs in terms of the g -good neighbor connectivity have been proposed, which provide a formula to compute the g -good-neighbor diagnosability for general graphs (therefore for Cartesian product graphs). As their applications, we get the exact values or bounds for the gc numbers and g -good-neighbor diagnosability of grid, torus networks and generalized cubes.}, archive = {J_IANDC}, author = {Zhao Wang and Yaping Mao and Sun-Yuan Hsieh and Ralf Klasing}, doi = {10.1016/j.ic.2025.105341}, journal = {Information and Computation}, month = {11}, pages = {105341}, shortjournal = {Inf. Comput.}, title = {The g-good-neighbor diagnosability of product networks under the PMC model}, volume = {307}, year = {2025}, }
• (2025). Two-state spin systems with negative interactions. IANDC, 307, 105340. (www)

  @article{ , abstract = {We study the approximability of computing the partition functions of two-state spin systems. The problem is parameterized by a 2 × 2 symmetric matrix. Previous results on this problem were restricted either to the case where the matrix has non-negative entries, or to the case where the diagonal entries are equal, i.e. Ising models. In this paper, we study the generalization to arbitrary 2 × 2 interaction matrices with real entries. We show that in some regions of the parameter space, it's #P-hard to even determine the sign of the partition function, while in other regions there are fully polynomial approximation schemes for the partition function. Our results reveal several new computational phase transitions.}, archive = {J_IANDC}, author = {Yumou Fei and Leslie Ann Goldberg and Pinyan Lu}, doi = {10.1016/j.ic.2025.105340}, journal = {Information and Computation}, month = {11}, pages = {105340}, shortjournal = {Inf. Comput.}, title = {Two-state spin systems with negative interactions}, volume = {307}, year = {2025}, }
• (2025). Competition among parallel contests. IANDC, 307, 105339. (www)

  @article{ , abstract = {We investigate the model of multiple rank-order contests held in parallel, where each contestant only selects one contest to join and each contest designer decides the prize structure to compete for the participation of contestants. We first analyze the strategic behaviors of contestants and completely characterize the symmetric Bayesian Nash equilibrium. As for the strategies of contest designers, when other designers' strategies are known, we show that computing the best response is NP-hard and propose a fully polynomial time approximation scheme to output the ϵ -approximate best response. When other designers' strategies are unknown, we provide a worst-case analysis on one designer's strategy. We give an upper bound on the worst-case utility of any strategy and propose a method to construct a strategy whose utility can guarantee a constant ratio of this upper bound in the worst case.}, archive = {J_IANDC}, author = {Xiaotie Deng and Ningyuan Li and Weian Li and Qi Qi}, doi = {10.1016/j.ic.2025.105339}, journal = {Information and Computation}, month = {11}, pages = {105339}, shortjournal = {Inf. Comput.}, title = {Competition among parallel contests}, volume = {307}, year = {2025}, }

ICV - 12

• (2025). Test-time adaptation for object detection via dynamic dual teaching. ICV, 163, 105740. (www)

  @article{ , abstract = {Test-Time Adaptation (TTA) is a practical setting in real-world applications, which aims to adapt a source-trained model to target domains with online unlabeled test data streams. Current approaches often rely on self-training, utilizing supervision signals from the source-trained model, suffering from poor adaptation due to diverse domain shifts. In this paper, we propose a novel test-time adaptation method for object detection guided by dual teachers, termed D ynamic D ual T eaching ( DDT ). Inspired by the generalization potentials of Vision-Language Models (VLMs), we integrate the VLM as an additional language-driven instructor. This integration exploits the domain-robustness of language prompts to mitigate domain shifts, collaborating with the teacher of source information within the teacher–student framework. Firstly, we utilize an ensemble prompt to guide the prediction process of the language-driven instructor. Secondly, a dynamic fusion strategy of the dual teachers is designed to generate high-quality pseudo-labels for student learning. Moreover, we incorporate a dual prediction consistency regularization to further mitigate the sensitivity of the adapted detector to domain shifts. Experiments on diverse domain adaptation benchmarks demonstrate that the proposed DDT method achieves state-of-the-art performance on both online and offline domain adaptation settings.}, archive = {J_ICV}, author = {Siqi Zhang and Lu Zhang and Zhiyong Liu}, doi = {10.1016/j.imavis.2025.105740}, journal = {Image and Vision Computing}, month = {11}, pages = {105740}, shortjournal = {Image Vis. Comput.}, title = {Test-time adaptation for object detection via dynamic dual teaching}, volume = {163}, year = {2025}, }
• (2025). Mining fine-grained attributes for vision–semantics integration in few-shot learning. ICV, 163, 105739. (www)

  @article{ , abstract = {Recent advancements in Few-Shot Learning (FSL) have been significantly driven by leveraging semantic descriptions to enhance feature discrimination and recognition performance. However, existing methods, such as SemFew, often rely on verbose or manually curated attributes and apply semantic guidance only to the support set, limiting their effectiveness in distinguishing fine-grained categories. Inspired by human visual perception, which emphasizes crucial features for accurate recognition, this study introduces concise, fine-grained semantic attributes to address these limitations. We propose a Visual Attribute Enhancement (VAE) mechanism that integrates enriched semantic information into visual features, enabling the model to highlight the most relevant visual attributes and better distinguish visually similar samples. This module enhances visual features by aligning them with semantic attribute embeddings through a cross-attention mechanism and optimizes this alignment using an attribute-based cross-entropy loss. Furthermore, to mitigate the performance degradation caused by methods that supply semantic information exclusively to the support set, we propose a semantic attribute reconstruction (SAR) module. This module predicts and integrates semantic features for query samples, ensuring balanced information distribution between the support and query sets. Specifically, SAR enhances query representations by aligning and reconstructing semantic and visual attributes through regression and optimal transport losses to ensure semantic–visual consistency. Experiments on five benchmark datasets, including both general datasets and more challenging fine-grained Few-Shot datasets consistently demonstrate that our proposed method outperforms state-of-the-art methods in both 5-way 1-shot and 5-way 5-shot settings.}, archive = {J_ICV}, author = {Juan Zhao and Lili Kong and Deshang Sun and Deng Xiong and Jiancheng Lv}, doi = {10.1016/j.imavis.2025.105739}, journal = {Image and Vision Computing}, month = {11}, pages = {105739}, shortjournal = {Image Vis. Comput.}, title = {Mining fine-grained attributes for vision–semantics integration in few-shot learning}, volume = {163}, year = {2025}, }
• (2025). Explainable deepfake detection across different modalities: An overview of methods and challenges. ICV, 163, 105738. (www)

  @article{ , abstract = {The increasing use of deepfake technology enables the creation of realistic and deceptive content, raising concerns about several serious issues, including biometric authentication, misinformation, politics, privacy, and trust. Many Deepfake Detection (DD) models are entering the market to combat the misuse of deepfakes. With these developments, one primary issue occurs in ensuring the explainability of the proposed detection models to understand the rationale of the decision. This paper aims to investigate the state-of-the-art explainable DD models across multiple modalities, including image, video, audio, and text. Unlike existing surveys that focus on detection methodologies with minimal attention to explainability and limited modality coverage, this paper directly focuses on these gaps. It offers a comprehensive analysis of advanced explainability techniques, including Grad-CAM, LIME, SHAP, LRP, Saliency Maps, and Anchors, for detecting deceptive content across the modalities. It identifies the strengths and limitations of existing models and outlines research directions to enhance explainability and interpretability in future works. By exploring these models, we aim to enhance transparency, provide deeper insights into model decisions, and bridge the gap between detection accuracy with explainability in DD models.}, archive = {J_ICV}, author = {MD Sarfaraz Momin and Abu Sufian and Debaditya Barman and Marco Leo and Cosimo Distante and Naser Damer}, doi = {10.1016/j.imavis.2025.105738}, journal = {Image and Vision Computing}, month = {11}, pages = {105738}, shortjournal = {Image Vis. Comput.}, title = {Explainable deepfake detection across different modalities: An overview of methods and challenges}, volume = {163}, year = {2025}, }
• (2025). MITS: A large-scale multimodal benchmark dataset for intelligent traffic surveillance. ICV, 163, 105736. (www)

  @article{ , abstract = {General-domain large multimodal models (LMMs) have achieved significant advances in various image-text tasks. However, their performance in the Intelligent Traffic Surveillance (ITS) domain remains limited due to the absence of dedicated multimodal datasets. To address this gap, we introduce MITS (Multimodal Intelligent Traffic Surveillance), the first large-scale multimodal benchmark dataset specifically designed for ITS. MITS includes 170,400 independently collected real-world ITS images sourced from traffic surveillance cameras, annotated with eight main categories and 24 subcategories of ITS-specific objects and events under diverse environmental conditions. Additionally, through a systematic data generation pipeline, we generate high-quality image captions and 5 million instruction-following visual question-answer pairs , addressing five critical ITS tasks : object and event recognition, object counting, object localization, background analysis, and event reasoning. To demonstrate MITS’s effectiveness, we fine-tune mainstream LMMs on this dataset, enabling the development of ITS-specific applications. Experimental results show that MITS significantly improves LMM performance in ITS applications, increasing LLaVA-1.5’s performance from 0.494 to 0.905 (+83.2%), LLaVA-1.6’s from 0.678 to 0.921 (+35.8%), Qwen2-VL’s from 0.584 to 0.926 (+58.6%), and Qwen2.5-VL’s from 0.732 to 0.930 (+27.0%). We release the dataset, code, and models as open-source , providing high-value resources to advance both ITS and LMM research.}, archive = {J_ICV}, author = {Kaikai Zhao and Zhaoxiang Liu and Peng Wang and Xin Wang and Zhicheng Ma and Yajun Xu and Wenjing Zhang and Yibing Nan and Kai Wang and Shiguo Lian}, doi = {10.1016/j.imavis.2025.105736}, journal = {Image and Vision Computing}, month = {11}, pages = {105736}, shortjournal = {Image Vis. Comput.}, title = {MITS: A large-scale multimodal benchmark dataset for intelligent traffic surveillance}, volume = {163}, year = {2025}, }
• (2025). UNIR-net: A novel approach for restoring underwater images with non-uniform illumination using synthetic data. ICV, 163, 105734. (www)

  @article{ , abstract = {Restoring underwater images affected by non-uniform illumination (NUI) is essential to improve visual quality and usability in marine applications. Conventional methods often fall short in handling complex illumination patterns, while learning-based approaches face challenges due to the lack of targeted datasets. To address these limitations, the Underwater Non-uniform Illumination Restoration Network (UNIR-Net) is proposed. UNIR-Net integrates multiple components, including illumination enhancement, attention mechanisms, visual refinement, and contrast correction, to effectively restore underwater images affected by NUI. In addition, the Paired Underwater Non-uniform Illumination (PUNI) dataset is introduced, specifically designed for training and evaluating models under NUI conditions. Experimental results on PUNI and the large-scale real-world Non-Uniform Illumination Dataset (NUID) show that UNIR-Net achieves superior performance in both quantitative metrics and visual outcomes. UNIR-Net also improves downstream tasks such as underwater semantic segmentation, highlighting its practical relevance. The code is available at https://github.com/xingyumex/UNIR-Net .}, archive = {J_ICV}, author = {Ezequiel Pérez-Zarate and Chunxiao Liu and Oscar Ramos-Soto and Diego Oliva and Marco Pérez-Cisneros}, doi = {10.1016/j.imavis.2025.105734}, journal = {Image and Vision Computing}, month = {11}, pages = {105734}, shortjournal = {Image Vis. Comput.}, title = {UNIR-net: A novel approach for restoring underwater images with non-uniform illumination using synthetic data}, volume = {163}, year = {2025}, }
• (2025). FFENet: A frequency fusion and enhancement network for camouflaged object detection. ICV, 163, 105733. (www)

  @article{ , abstract = {The goal of camouflaged object detection (COD) is to accurately find camouflaged objects hidden in their surroundings. Although most of the existing frequency-domain based COD models can boost the performance of COD to a certain extent by utilizing the frequency domain information, the frequency feature fusion strategies they adopt tend to ignore the complementary effects between high-frequency features and low-frequency features. In addition, most of the existing frequency-domain based COD models also do not consider enhancing camouflaged objects using low-level frequency-domain features. In order to solve these problems, we present a frequency fusion and enhancement network (FFENet) for camouflaged object detection, which mainly includes three stages. In the frequency feature extraction stage, we design a frequency feature learning module (FLM) to extract corresponding high-frequency features and low-frequency features. In the frequency feature fusion stage, we design a frequency feature fusion module (FFM) that can increase the representation ability of the fused features by adaptively assigning weights to the high-frequency features and the low-frequency features using a cross-attention mechanism. In the frequency feature guidance information enhancement stage, we design a frequency feature guidance information enhancement module (FGIEM) to enhance the contextual information and detail information of camouflaged objects in the fused features under the guidance of the low-level frequency features. Extensive experimental results on the COD10K, CHAMELEON, NC4K and CAMO datasets show that our model is superior to most existing COD models.}, archive = {J_ICV}, author = {Haishun Du and Wenzhe Zhang and Sen Wang and Zhengyang Zhang and Linbing Cao}, doi = {10.1016/j.imavis.2025.105733}, journal = {Image and Vision Computing}, month = {11}, pages = {105733}, shortjournal = {Image Vis. Comput.}, title = {FFENet: A frequency fusion and enhancement network for camouflaged object detection}, volume = {163}, year = {2025}, }
• (2025). UpAttTrans: Upscaled attention based transformer for facial image super-resolution. ICV, 163, 105731. (www)

  @article{ , abstract = {Image super-resolution (SR) aims to reconstruct high-quality images from low-resolution inputs, a task particularly challenging in face-related applications due to extreme degradations and modality differences (e.g., visible, low-resolution, near-infrared). Conventional convolutional neural networks (CNNs) and GAN-based approaches have achieved notable success; however, they often struggle with preserving identity and fine structural details at high upscaling factors. In this work, we introduce UpAttTrans, a novel attention mechanism that connects original and upsampled features for better detail recovery based on vision transformer for SR. The core generator leverages a custom UpAttTrans module that translates input image patches into embeddings, processes them through transformer layers enhanced with connector-up attention, and reconstructs high-resolution outputs with improved detail retention. We evaluate our model on the CelebA dataset across multiple upscaling factors ( 4 × , 8 × , 16 × , 32 × , and 64 × ). UpAttTrans achieves a 24.63% increase in PSNR, 21.56% in SSIM, and 19.61% reduction in FID for 4 × and 8 × SR, outperforming state-of-the-art baselines. Additionally, for higher magnification levels, our model maintains strong performance, with average gains of 6.20% in PSNR and 21.49% in SSIM, indicating its robustness in extreme SR settings. These findings suggest that UpAttTrans holds significant promise for real-world applications such as face recognition in surveillance, forensic image enhancement, and cross-spectral matching, where high-quality reconstruction from severely degraded inputs is critical.}, archive = {J_ICV}, author = {Neeraj Baghel and Shiv Ram Dubey and Satish Kumar Singh}, doi = {10.1016/j.imavis.2025.105731}, journal = {Image and Vision Computing}, month = {11}, pages = {105731}, shortjournal = {Image Vis. Comput.}, title = {UpAttTrans: Upscaled attention based transformer for facial image super-resolution}, volume = {163}, year = {2025}, }
• (2025). Novel extraction of discriminative fine-grained feature to improve retinal vessel segmentation. ICV, 163, 105729. (www)

  @article{ , abstract = {Retinal vessel segmentation is a vital early detection method for several severe ocular diseases. Despite significant progress in retinal vessel segmentation with the advancement of Neural Networks, there are still challenges to overcome. Specifically, retinal vessel segmentation aims to predict the class label for every pixel within a fundus image, with a primary focus on intra-image discrimination, making it vital for models to extract more discriminative features. Nevertheless, existing methods primarily focus on minimizing the difference between the output from the decoder and the label, but ignore fully using feature-level fine-grained representations from the encoder. To address these issues, we propose a novel Attention U-shaped Kolmogorov–Arnold Network named AttUKAN along with a novel Label-guided Pixel-wise Contrastive Loss for retinal vessel segmentation. Specifically, we implement Attention Gates into Kolmogorov–Arnold Networks to enhance model sensitivity by suppressing irrelevant feature activations and model interpretability by non-linear modeling of KAN blocks. Additionally, we also design a novel Label-guided Pixel-wise Contrastive Loss to supervise our proposed AttUKAN to extract more discriminative features by distinguishing between foreground vessel-pixel pairs and background pairs. Experiments are conducted across four public datasets including DRIVE, STARE, CHASE_DB1, HRF and our private dataset. AttUKAN achieves F1 scores of 82.50%, 81.14%, 81.34%, 80.21% and 80.09%, along with MIoU scores of 70.24%, 68.64%, 68.59%, 67.21% and 66.94% in the above datasets, which are the highest compared to 11 networks for retinal vessel segmentation. Quantitative and qualitative results show that our AttUKAN achieves state-of-the-art performance and outperforms existing retinal vessel segmentation methods. Our code will be available at https://github.com/stevezs315/AttUKAN .}, archive = {J_ICV}, author = {Shuang Zeng and Chee Hong Lee and Micky C. Nnamdi and Wenqi Shi and J. Ben Tamo and Hangzhou He and Xinliang Zhang and Qian Chen and May D. Wang and Lei Zhu and Yanye Lu and Qiushi Ren}, doi = {10.1016/j.imavis.2025.105729}, journal = {Image and Vision Computing}, month = {11}, pages = {105729}, shortjournal = {Image Vis. Comput.}, title = {Novel extraction of discriminative fine-grained feature to improve retinal vessel segmentation}, volume = {163}, year = {2025}, }
• (2025). Artificial intelligence content detection techniques using watermarking: A survey. ICV, 163, 105728. (www)

  @article{ , abstract = {The rapid advancement in AI-generated content has catalyzed artistic creation, advertising, and media dissemination. Despite their widespread applications across several domains, AI-generated content inherently poses risks of identity fraud, copyright violation and unauthorized use. Watermarking has emerged as a critical tool for copyright protection, allowing embedding of identification information in AI-generated content, and enhances traceability and verification without hurting user experience. In this study, we provide a systematic literature review of the technique for detecting AI content, especially text and images, using watermarking, spanning studies from 2010 to 2025. Studies included in this review were peer-reviewed articles that applied watermarking to effectively distinguish AI-generated content from real or human-written content. We report strong past and current approaches to detecting watermarking-based AI content, especially text and images. This includes an analysis of how watermarking methods are used on AI-generated content, their role in enhancing performance, and a detail comparative analysis of notable techniques. Furthermore, we discuss how these methods have been evaluated, identify the research gaps and potential solutions. Our findings provide valuable insights for future watermarking-based AI content detection researchers, applications and organizations seeking to implement watermarking solutions in potential applications. To the best of our knowledge, we are the first to explore the detection of AI content, especially text and image, detection using watermarking.}, archive = {J_ICV}, author = {Nishant Kumar and Amit Kumar Singh}, doi = {10.1016/j.imavis.2025.105728}, journal = {Image and Vision Computing}, month = {11}, pages = {105728}, shortjournal = {Image Vis. Comput.}, title = {Artificial intelligence content detection techniques using watermarking: A survey}, volume = {163}, year = {2025}, }
• (2025). Your image generator is your new private dataset. ICV, 163, 105727. (www)

  @article{ , abstract = {Generative diffusion models have emerged as powerful tools to synthetically produce training data, offering potential solutions to data scarcity and reducing labelling costs for downstream supervised deep learning applications. However, existing approaches for synthetic dataset generation face significant limitations: previous methods like Knowledge Recycling rely on label-conditioned generation with models trained from scratch, limiting flexibility and requiring extensive computational resources, while simple class-based conditioning fails to capture the semantic diversity and intra-class variations found in real datasets. Additionally, effectively leveraging text-conditioned image generation for building classifier training sets requires addressing key issues: constructing informative textual prompts, adapting generative models to specific domains, and ensuring robust performance. This paper proposes the Text-Conditioned Knowledge Recycling (TCKR) pipeline to tackle these challenges. TCKR combines dynamic image captioning, parameter-efficient diffusion model fine-tuning, and Generative Knowledge Distillation techniques to create synthetic datasets tailored for image classification. The pipeline is rigorously evaluated on ten diverse image classification benchmarks. The results demonstrate that models trained solely on TCKR-generated data achieve classification accuracies on par with (and in several cases exceeding) models trained on real images. Furthermore, the evaluation reveals that these synthetic-data-trained models exhibit substantially enhanced privacy characteristics: their vulnerability to Membership Inference Attacks is significantly reduced, with the membership inference AUC lowered by 5.49 points on average compared to using real training data, demonstrating a substantial improvement in the performance-privacy trade-off. These findings indicate that high-fidelity synthetic data can effectively replace real data for training classifiers, yielding strong performance whilst simultaneously providing improved privacy protection as a valuable emergent property. The code and trained models are available in the accompanying open-source repository .}, archive = {J_ICV}, author = {Nicolò Francesco Resmini and Eugenio Lomurno and Cristian Sbrolli and Matteo Matteucci}, doi = {10.1016/j.imavis.2025.105727}, journal = {Image and Vision Computing}, month = {11}, pages = {105727}, shortjournal = {Image Vis. Comput.}, title = {Your image generator is your new private dataset}, volume = {163}, year = {2025}, }
• (2025). Assessing the noise robustness of class activation maps: A framework for reliable model interpretability. ICV, 163, 105717. (www)

  @article{ , abstract = {Class Activation Maps (CAMs) are one of the important methods for visualizing regions used by deep learning models. Yet their robustness to different noise remains underexplored. In this work, we evaluate and report the resilience of various CAM methods for different noise perturbations across multiple architectures and datasets. By analyzing the influence of different noise types on CAM explanations, we assess the susceptibility to noise and the extent to which dataset characteristics may impact explanation stability. The findings highlight considerable variability in noise sensitivity for various CAMs. We propose a robustness metric for CAMs that captures two key properties: consistency and responsiveness. Consistency reflects the ability of CAMs to remain stable under input perturbations that do not alter the predicted class, while responsiveness measures the sensitivity of CAMs to changes in the prediction caused by such perturbations. The metric is evaluated empirically across models, different perturbations, and datasets along with complementary statistical tests to exemplify the applicability of our proposed approach.}, archive = {J_ICV}, author = {Syamantak Sarkar and Revoti P. Bora and Bhupender Kaushal and Sudhish N. George and Kiran Raja}, doi = {10.1016/j.imavis.2025.105717}, journal = {Image and Vision Computing}, month = {11}, pages = {105717}, shortjournal = {Image Vis. Comput.}, title = {Assessing the noise robustness of class activation maps: A framework for reliable model interpretability}, volume = {163}, year = {2025}, }
• (2025). InceptionWTMNet: A hybrid network for alzheimer’s disease detection using wavelet transform convolution and mixed local channel attention on finely fused multimodal images. ICV, 163, 105693. (www)

  @article{ , abstract = {Multimodal fusion has emerged as a critical technique for the diagnosis of Alzheimer’s Disease (AD), with the aim of effectively extracting and utilising complementary information from diverse modalities. Current fusion methods frequently cause the precise alignment of source images and do not adequately address parallax issues. This oversight can result in artifacts during the fusion process when images are misaligned. In response to this challenge, we propose a refined registration fusion technique, termed MURF, which integrates multimodal image registration and fusion within a cohesive framework. The Vision Transformer (ViT) has inspired the application of large-kernel convolutions in the diagnosis of Alzheimer’s disease (AD) because of its ability to model long-range dependencies. This approach aims to expand the receptive field and enhance the performance of diagnostic models. Despite requiring a minimal number of floating-point operations (FLOPs), these deep operators encounter challenges associated with over-parameterisation because of high memory access costs, which ultimately compromises computational efficiency. By utilising wavelet transform convolutions (WTConv), we decompose large-kernel depth-wise convolutions into four parallel branches. One branch employs a wavelet-transform convolution with square kernels, while the other two branches incorporate orthogonal wavelet-transform kernels with an identity mapping. This innovative method, with a Mixed Local Channel Attention mechanism, has facilitated the development of the InceptionWTConvolutions network. This network maintains a receptive field comparable to that of large-kernel convolutions, while concurrently minimising over-parameterisation and enhancing computational efficiency. InceptionWTMNet classified AD, MCI, and NC using MRI and PET data from ADNI dataset with 98.69% accuracy, 98.65% recall, 98.70% F1-score, and 98.98% AUC. and provide Graphical abstract in correct format.}, archive = {J_ICV}, author = {Zenan Xu and Zhengyao Bai and Han Ma and Mingqiang Xu and Qiqin Huang and Tao Lin}, doi = {10.1016/j.imavis.2025.105693}, journal = {Image and Vision Computing}, month = {11}, pages = {105693}, shortjournal = {Image Vis. Comput.}, title = {InceptionWTMNet: A hybrid network for alzheimer’s disease detection using wavelet transform convolution and mixed local channel attention on finely fused multimodal images}, volume = {163}, year = {2025}, }

IJAR - 1

• (2026). Choosing the center of star-shaped set-valued data compatible with measure-preserving arithmetic. IJAR, 188, 109575. (www)

  @article{ , abstract = {Set-valued data has traditionally been represented by considering non-empty compact and convex subsets of R d with the usual Minkowski addition. An alternative and flexible setting that admits a functional representation are the star-shaped sets. A framework based on a center-radial characterization has been introduced to treat these sets from a statistical point of view. The arithmetic is defined directionally, which is more natural for representing imprecision propagation in higher dimensions. Nevertheless, the problem of determining a center for star-shaped sets coherent with the arithmetic and sound for statistical purposes has not been fully addressed yet. The aim is to advance on the directional characterization for star-shaped sets by considering a measure-preserving arithmetic together with a center selection fully compatible with this arithmetic. The practicability of the new framework will be illustrated using a classical dataset in set-valued statistics.}, archive = {J_IJAR}, author = {Gil González-Rodríguez}, doi = {10.1016/j.ijar.2025.109575}, journal = {International Journal of Approximate Reasoning}, month = {1}, pages = {109575}, shortjournal = {Int. J. Approx. Reasoning}, title = {Choosing the center of star-shaped set-valued data compatible with measure-preserving arithmetic}, volume = {188}, year = {2026}, }

IPL - 10

• (2026). Tighter bounds on non-clairvoyant parallel machine scheduling with prediction to minimize makespan. IPL, 191, 106598. (www)

  @article{ , abstract = {This paper investigates the non-clairvoyant parallel machine scheduling problem with prediction, with the objective of minimizing the makespan. Improved lower bounds for the problem and competitive ratios of online algorithms with respect to the prediction error are presented for both the non-preemptive and preemptive cases on m identical machines.}, archive = {J_IPL}, author = {Tianqi Chen and Zhiyi Tan}, doi = {10.1016/j.ipl.2025.106598}, journal = {Information Processing Letters}, month = {1}, pages = {106598}, shortjournal = {Inf. Process. Lett.}, title = {Tighter bounds on non-clairvoyant parallel machine scheduling with prediction to minimize makespan}, volume = {191}, year = {2026}, }
• (2026). Note on pancake sorting. IPL, 191, 106597. (www)

  @article{ , abstract = {We present generalized approach to the proof of the lower bound for unburnt pancake sorting problem, where we search for the number f ( n ) of prefix reversals required to sort a stack (permutation) of n pancakes. For this purpose we introduce a new concept of guarded pancake blocks. Gates and Papadimitriou proved that f ( n ) ≥ 17 n / 16 for n a multiple of 16. Heydari and Sudborough improved this bound to f ( n ) ≥ 15 n / 14 for n a multiple of 14. We extend that result to f ( n ) ≥ ⌊ ( 15 n + 9 ) / 14 ⌋ for every n ≥ 6 .}, archive = {J_IPL}, author = {Marcin Peczarski}, doi = {10.1016/j.ipl.2025.106597}, journal = {Information Processing Letters}, month = {1}, pages = {106597}, shortjournal = {Inf. Process. Lett.}, title = {Note on pancake sorting}, volume = {191}, year = {2026}, }
• (2026). Logical characterization of branching bisimilarity over random processes. IPL, 191, 106596. (www)

  @article{ , abstract = {Quantitative aspects like probabilities play an important role in concurrent processes. Providing a (modal) logic for a randomized concurrency model can augment the toolbox of analyzing probabilistic processes, and thus is a frequent topic in the field. In this paper, we are interested in logically characterizing uniformly randomized processes, whose semantical behavior is defined in a model-independent manner. Specifically, we present two modal logics for the uniformly randomized version of finite-state CCS (RCCS fs for short). Our logics extend the Hennessy-Milner logic, and one of them is equipped with the μ operator. Indeed, we prove that both logics characterize the branching bisimilarity for RCCS fs , i.e., two RCCS fs processes are branching bisimilar if and only if they are logically equivalent. To facilitate the proof, we also develop for RCCS fs an up-to proof method, which may be of independent interest.}, archive = {J_IPL}, author = {Xian Xu and Wenbo Zhang}, doi = {10.1016/j.ipl.2025.106596}, journal = {Information Processing Letters}, month = {1}, pages = {106596}, shortjournal = {Inf. Process. Lett.}, title = {Logical characterization of branching bisimilarity over random processes}, volume = {191}, year = {2026}, }
• (2026). Notes about the linear complexity of quaternary cyclotomic sequences of order four. IPL, 191, 106595. (www)

  @article{ , abstract = {In this paper, we study the linear complexity of quaternary cyclotomic sequences with period p , where p > 2 is a prime. Considered sequences are based on classical cyclotomic classes of order four modulo p . We show that any balanced quaternary cyclotomic sequence of order four with period p has high linear complexity over finite ring of order four. Our results generalize those obtained earlier by other authors.}, archive = {J_IPL}, author = {Vladimir Edemskiy and Zeyu Cao}, doi = {10.1016/j.ipl.2025.106595}, journal = {Information Processing Letters}, month = {1}, pages = {106595}, shortjournal = {Inf. Process. Lett.}, title = {Notes about the linear complexity of quaternary cyclotomic sequences of order four}, volume = {191}, year = {2026}, }
• (2026). Finding the cyclic covers of a string. IPL, 191, 106594. (www)

  @article{ , abstract = {We introduce the concept of cyclic covers, which generalizes the classical notion of covers in strings. Given any string X , a factor W of X is called a cyclic cover if each position of X belongs to an occurrence of a cyclic shift of W in X . Two cyclic covers are distinct if one is not a cyclic shift of the other. The cyclic covers problem asks for all distinct cyclic covers of an input string X . We present an algorithm that solves the cyclic covers problem in O ( n log ⁡ n ) time, where n is the length of X . It is based on finding a well-structured set of standard occurrences of a constant number of factors of a cyclic cover candidate W , computing the regions of X covered by cyclic shifts of W , extending those factors, and taking the union of the results.}, archive = {J_IPL}, author = {Roberto Grossi and Costas S. Iliopoulos and Jesper Jansson and Zara Lim and Wing-Kin Sung and Wiktor Zuba}, doi = {10.1016/j.ipl.2025.106594}, journal = {Information Processing Letters}, month = {1}, pages = {106594}, shortjournal = {Inf. Process. Lett.}, title = {Finding the cyclic covers of a string}, volume = {191}, year = {2026}, }
• (2026). Rank-2 module-LIP with special matrices. IPL, 191, 106593. (www)

  @article{ , abstract = {Lattice isomorphism problem (LIP) has been studied since 1990s. In 2023, a post-quantum signature scheme known as HAWK was submitted in the NIST standardization of additional signature scheme, which is based on the module lattice isomorphism problem (module-LIP). Module-LIP was formally defined by Mureau et al. at Eurocrypt'24 and Luo et al. reduced the problem of solving module-LIP over CM number fields to a problem of finding the special type of symplectic automorphism. In this paper, we extend this idea further by establishing a reduction of the module-LIP to a problem of finding special types of matrices.}, archive = {J_IPL}, author = {Manoj Gyawali}, doi = {10.1016/j.ipl.2025.106593}, journal = {Information Processing Letters}, month = {1}, pages = {106593}, shortjournal = {Inf. Process. Lett.}, title = {Rank-2 module-LIP with special matrices}, volume = {191}, year = {2026}, }
• (2026). Corrigendum to “On the complexity of co-secure dominating set problem” [Inf. process. lett. 185 (2024) 106463]. IPL, 191, 106592. (www)

  @article{ , abstract = {We correct an error in Theorem 4 in our published paper Panda et al. [3] .}, archive = {J_IPL}, author = {Bhawani Sankar Panda and Soumyashree Rana and Sounaka Mishra}, doi = {10.1016/j.ipl.2025.106592}, journal = {Information Processing Letters}, month = {1}, pages = {106592}, shortjournal = {Inf. Process. Lett.}, title = {Corrigendum to “On the complexity of co-secure dominating set problem” [Inf. process. lett. 185 (2024) 106463]}, volume = {191}, year = {2026}, }
• (2026). On some families of binary codes. IPL, 191, 106591. (www)

  @article{ , abstract = {We give further results on the weight distributions of the two families of binary codes recently constructed by simplicial complexes by (Wu, Lee, 2020), and show that the converse of the above results is also correct, that is, the binary codes with such weight distributions properties must be these two families of codes. Based on the above results, we also construct another family of binary self-orthogonal codes and present their separating properties and applications to the secret sharing scheme, cryptography and other aspects of information security.}, archive = {J_IPL}, author = {Zihui Liu}, doi = {10.1016/j.ipl.2025.106591}, journal = {Information Processing Letters}, month = {1}, pages = {106591}, shortjournal = {Inf. Process. Lett.}, title = {On some families of binary codes}, volume = {191}, year = {2026}, }
• (2026). The complexity of computing the period and the exponent of a digraph. IPL, 191, 106590. (www)

  @article{ , abstract = {The period of a strongly connected digraph is the greatest common divisor of the lengths of all its cycles. The period of a digraph is the least common multiple of the periods of its strongly connected components. These notions play an important role in the theory of Markov chains and the analysis of powers of nonnegative matrices. While the time complexity of computing the period is well-understood, little is known about its space complexity. We show that the problem of computing the period of a digraph is NL -complete, even if all its cycles are contained in the same strongly connected component. However, if the digraph is strongly connected, we show that this problem becomes L -complete. For primitive digraphs (that is, strongly connected digraphs of period one), there always exists a number m such that there is a path of length exactly m between every two vertices. We show that computing the smallest such m , called the exponent of a digraph, is NL -complete. The exponent of a primitive digraph is a particular case of the index of convergence of a nonnegative matrix, which we also show to be computable in NL , and thus NL -complete.}, archive = {J_IPL}, author = {Stefan Kiefer and Andrew Ryzhikov}, doi = {10.1016/j.ipl.2025.106590}, journal = {Information Processing Letters}, month = {1}, pages = {106590}, shortjournal = {Inf. Process. Lett.}, title = {The complexity of computing the period and the exponent of a digraph}, volume = {191}, year = {2026}, }
• (2026). A simple supercritical tradeoff between size and height in resolution. IPL, 191, 106589. (www)

  @article{ , abstract = {We describe CNFs in n variables which, over a range of parameters, have small resolution refutations but are such that any small refutation must have height larger than n (even exponential in n ), where the height of a refutation is the length of the longest path in it. This is called a supercritical tradeoff between size and height because, if we do not care about size, every CNF is refutable in height n . Our proof method uses a simple construction, based on or-ification and base d representations of integers, to reduce the number of variables. A similar result appeared in [Fleming, Pitassi and Robere, ITCS '22], for different formulas using a more complicated construction for reducing the number of variables. Small refutations of our formula are necessarily highly irregular, making it a plausible candidate to separate resolution from pool resolution, which amounts to separating CDCL with restarts from CDCL without restarts. We are not able to show this. In the other direction, we show that a simpler version of our formula, with a similar irregularity property, does have polynomial size pool resolution refutations and thus does not provide such a separation for CDCL.}, archive = {J_IPL}, author = {Sam Buss and Neil Thapen}, doi = {10.1016/j.ipl.2025.106589}, journal = {Information Processing Letters}, month = {1}, pages = {106589}, shortjournal = {Inf. Process. Lett.}, title = {A simple supercritical tradeoff between size and height in resolution}, volume = {191}, year = {2026}, }

ISAT - 39

• (2025). Voltage tracking and regulation of vehicle PEMFC system under low load condition based on fuzzy LQG hybrid strategy. ISAT, 165, 510-523. (www)

  @article{ , abstract = {In automotive fuel cell systems, high-voltage operation accelerates carbon support and platinum catalyst degradation, significantly compromising system durability. This study develops a dynamic system model with active cathode recirculation to capture the transient response of voltage, and proposes a hybrid control scheme that combines a proportional compensator with a fuzzy LQG controller to effectively enhance voltage regulation and disturbance tracking capabilities. Extensive simulation and hardware-in-the-loop (HiL) confirm the precision and rapid response of the developed controller. Compared to single LQG and fuzzy LQG controllers, the error reduction achieved is 49.3 % and 40.3 %, respectively, and the overall control benefit ratio improves by 19.2 % and 11 %. This method balances dynamic response with control efforts, effectively reducing the risk of high voltage-induced degradation under low-load conditions.}, archive = {J_ISAT}, author = {Ze Liu and Sichuan Xu and Baitao Zhang and Sida Guo}, doi = {10.1016/j.isatra.2025.06.008}, journal = {ISA Transactions}, month = {10}, pages = {510-523}, shortjournal = {ISA Trans.}, title = {Voltage tracking and regulation of vehicle PEMFC system under low load condition based on fuzzy LQG hybrid strategy}, volume = {165}, year = {2025}, }
• (2025). Cascade alarm systems: A study on singular value analysis. ISAT, 165, 497-509. (www)

  @article{ , abstract = {This paper introduces a novel alarm system designed for use with both Independent and Identically Distributed (IID) and non-IID variables. The proposed algorithm, termed the Cascade Alarm System (CAS), utilizes the largest singular value of the signal as the basis for fault detection, employing k alarm subsystems. The greatest singular values extracted from the Lagged Covariance Matrix (LCM) of a sliding window constitute the output of the first alarm subsystem. The CAS offers two primary advantages. First, each subsystem independently generates its own alarm signal, resulting in a more flexible multilevel architecture. Second, the multilevel structure, founded on singular value decomposition (SVD), exhibits a filtering property that enhances its resilience to noise and inaccuracies. The maximum singular value effectively captures the essential information of the signal, ensuring that the filtering capabilities of the proposed method do not significantly compromise the performance of the alarm system or the integrity of critical signal information. The experimental results from the implementation of the proposed alarm system under various fault conditions demonstrate satisfactory performance. Additionally, the performance of the Cascade Alarm System has been compared with leading contemporary alarm system design methodologies, including median, moving average filters, delay timers, Cumulative Sum Control Chart (CUSUM), and serial method.}, archive = {J_ISAT}, author = {J. Taheri-Kalani and M. Aliyari-Shoorehdeli and Gh. Latif-Shabgahi}, doi = {10.1016/j.isatra.2025.06.023}, journal = {ISA Transactions}, month = {10}, pages = {497-509}, shortjournal = {ISA Trans.}, title = {Cascade alarm systems: A study on singular value analysis}, volume = {165}, year = {2025}, }
• (2025). Flow pulsation compensation based composite adaptive active disturbance rejection control for electro-hydrostatic actuators. ISAT, 165, 486-496. (www)

  @article{ , abstract = {Due to the residual pressure accumulated during the reciprocating movement of the plunger pump, there is a deviation between the flow pulsation and the theoretical calculation value. Current nonlinear control methods for the electro-hydrostatic actuator (EHA) often oversimplify and compensate for flow pulsation linearly, neglecting its nonlinear characteristics and deviation effects. This approach increases matching uncertainties and amplifies noise due to higher control gains, thus limiting the improvement of control performance. To address this issue, this paper proposes a composite adaptive disturbance rejection control method based on flow pulsation compensation for the EHA. This method equates the flow pulsation model of the pump to a combination of a theoretical flow pulsation control input term and a bounded disturbance term (the difference between the theoretical and actual flow pulsation), followed by the design of a composite adaptive law to handle parameter uncertainties, and the design of the expanded state observers based on position and pressure signals to estimate and compensate for the uncertainties nonlinearly. Finally, the effectiveness of the proposed method is verified by comparing with other control methods through experiments.}, archive = {J_ISAT}, author = {Yaowen Ge and Xiaowei Yang and Weilin Zhu and Wenxiang Deng and Jianyong Yao}, doi = {10.1016/j.isatra.2025.06.014}, journal = {ISA Transactions}, month = {10}, pages = {486-496}, shortjournal = {ISA Trans.}, title = {Flow pulsation compensation based composite adaptive active disturbance rejection control for electro-hydrostatic actuators}, volume = {165}, year = {2025}, }
• (2025). Polytopic inclusion-based model predictive control for quasi-LPV systems using vertex system models and gain scheduling. ISAT, 165, 474-485. (www)

  @article{ , abstract = {This paper proposes a Model Predictive Control (MPC) strategy for a class of Quasi-Linear Parameter-Varying (quasi-LPV) systems characterized by a measurable time-varying parameter. The core of the proposed quasi-LPV-MPC controller lies in the utilization of a polytopic representation along with a gain-scheduled controller. A terminal cost that depends explicitly on the scheduling parameter is used. However, for the implementation, a complementary cost function is used to frame the optimization problem at each vertex level so that the requirement of updating the varying parameters over the prediction horizon is relaxed. Though the resulting suboptimal controller involves more computational burden, the proposed method demonstrates improvement in control performance over traditional MPC schemes. Experimental validation on a cascaded coupled tank system underscores the practical efficacy of the proposed quasi-LPV-MPC controller, while simulation studies on a twin rotor multi-input multi-output system serve as an additional demonstration example case. Comparative performance evaluations against both linear and nonlinear MPCs clearly illustrate that the quasi-LPV-MPC offers better control precision, adaptability, and the overall system responsiveness, thus positioning it as an effective solution for quasi-LPV systems.}, archive = {J_ISAT}, author = {Rangoli Singh and Sandip Ghosh and Devender Singh}, doi = {10.1016/j.isatra.2025.05.051}, journal = {ISA Transactions}, month = {10}, pages = {474-485}, shortjournal = {ISA Trans.}, title = {Polytopic inclusion-based model predictive control for quasi-LPV systems using vertex system models and gain scheduling}, volume = {165}, year = {2025}, }
• (2025). Research on the global energy optimization of multi-source and multi-actuator hydraulic systems based on dynamic programming and improved adaptive genetic algorithm. ISAT, 165, 450-473. (www)

  @article{ , abstract = {Multi-source and multi-actuator hydraulic systems (MSAHSs) are widely used in high-power energy transmission and construction machinery. However, individual control of each component without considering the overall power matching leads the system to the low-efficiency zone, results in environmental pollution and huge economic loss. Therefore, it is highly desirable to find a way of obtaining energy-saving green MSAHSs. In this paper, the power consumption model of closed MSAHSs is established firstly to analyze theoretical factors affecting the component efficiency and find that the hydraulic pressure is the key factor. On this basis, a multi-algorithm integration global power matching method is then proposed, which consist of back propagation (BP) neural network, dynamic programming (DP) and improved adaptive genetic algorithm (IAGA). BP is used to construct efficiency prediction models for power elements (pumps, motors and engines) respectively, DP is used for elements’ high efficiency zone preliminary search, and IAGA is used to realize the global power matching of the multiple power units with energy conversion and transfer finally through optimal control parameters precise searching. Experiment is conducted on the closed MSAHS in a hydraulic fracturing vehicle. Results demonstrate that the MSAHS applied with multi-algorithm integration method improves the overall efficiency to a highest fuel savings of 35.5 % under normal conditions compared with local power matching control.}, archive = {J_ISAT}, author = {Yuhang Zhong and Wenting Chen and Zihao Chen and Guanyu Zhai and Chao Ai and Gexin Chen}, doi = {10.1016/j.isatra.2025.06.010}, journal = {ISA Transactions}, month = {10}, pages = {450-473}, shortjournal = {ISA Trans.}, title = {Research on the global energy optimization of multi-source and multi-actuator hydraulic systems based on dynamic programming and improved adaptive genetic algorithm}, volume = {165}, year = {2025}, }
• (2025). Intelligent energy adaptive control of loader shoveling system. ISAT, 165, 437-449. (www)

  @article{ , abstract = {Loaders are often faced with various working objects during the shoveling process. The differences in working resistance and its time-varying unpredictability when shoveling different objects are the main causes of high energy consumption during the shoveling stage. In this paper, through the analysis of the shoveling process, the influence of the compacted layer on the working resistance is obtained. The constructed Discrete Element Method (DEM) simulation model is used to elucidate that the timely lifting of the boom can have a destructive effect on the compacted layer. Moreover, considering the diversity of working objects, a study was carried out on the effect of different boom lifting ranges on the destruction of the compacted layer. The loader shoveling system's intelligent Energy Adaptive Control (EAC) strategy is constructed by integrating the material recognition model based on the Back Propagation (BP) neural network algorithm. This control strategy can output the set pilot pressure according to the material type, realize the intelligent adjustment of the lifting range of the boom with the change of material type, and reduce the working resistance during the shoveling stage. The peak engine power consumed while shoveling sand, gravel, and boulders decreased by 20.6 %, 19.1 %, and 10.9 %, respectively, improving the energy utilization rate of the loader shoveling system when facing different working objects.}, archive = {J_ISAT}, author = {Bingwei Cao and Changhao Mu and Jiaqi Dong and Guangliang Tian and Yuqi Wang}, doi = {10.1016/j.isatra.2025.06.021}, journal = {ISA Transactions}, month = {10}, pages = {437-449}, shortjournal = {ISA Trans.}, title = {Intelligent energy adaptive control of loader shoveling system}, volume = {165}, year = {2025}, }
• (2025). Analysis and design of oscillation frequency correction for servo resonance suppression. ISAT, 165, 422-436. (www)

  @article{ , abstract = {Mechanical resonance poses significant hazards to the normal operation of the servo systems. To mitigate mechanical resonance, online adaptive notch filter is extensive used, thus the precise determination of resonant frequency holds significant importance. However, in certain scenarios involving the high-bandwidth servo system, a phenomenon known as frequency shift can make the notch filter ineffective in addressing servo resonance. To solve this problem, an oscillation frequency correction scheme based on two improved sliding-mode observers (ISMOs) utilizing a dual-power approximation law is proposed. First, the oscillation frequency shift is analyzed around the system delay, which can be equivalently modeled using a Pade approximation method. Subsequently, a feedback loop featuring two adaptive feedback coefficients is designed to automatically tune the time factor. Remarkably, the scheme can dynamically correct oscillation frequency, thereby promoting resonance suppression. At the same time, ISMOs-identified mechanical parameters provide critical foundations for feedback coefficient adjustment. It is worth noting that the dual-power approximation law effectively suppresses high-frequency chatter while maintaining parameter identification accuracy. Finally, the effectiveness of the scheme is validated through simulation and experimental results.}, archive = {J_ISAT}, author = {Yanan Tang and Shaowu Lu and Puliang Yu and Bao Song}, doi = {10.1016/j.isatra.2025.06.011}, journal = {ISA Transactions}, month = {10}, pages = {422-436}, shortjournal = {ISA Trans.}, title = {Analysis and design of oscillation frequency correction for servo resonance suppression}, volume = {165}, year = {2025}, }
• (2025). A segmented model based internal model control scheme of electromagnetic micro-mirror systems. ISAT, 165, 408-421. (www)

  @article{ , abstract = {In this paper, a segmented internal model control (SIMC) scheme based on the segmented combination model of the electromagnetic micro-mirror system (EMMS) is established. Notice that highly underdamped oscillation and rate-dependent hysteresis exist in the EMMS, so it is a complex nonlinear dynamic system. In order to control the deflection angle of the EMMS using the internal model control strategy, it is necessary to establish the inverse model of the EMMS. Therefore, a new model structure that is convenient for inversion is proposed in this paper to describe the characteristics of the EMMS with underdamped and rate-dependent hysteresis. In the proposed scheme, the model is a combination of a group of weighted sub-models based on the segmentation of the system's operating frequency. The weight of each segmented sub-model is not a constant but a new type of function which is also called the smoothing factor. Its function is to smooth the switching between sub-models, thereby reducing the dynamic error caused by model switching. In addition, the particle swarm optimization (PSO) algorithm is used to determine the optimal frequency segmentation points, which helps to obtain the optimal model for describing the system characteristics. Based on the proposed segmented combination model, the corresponding segmented internal model control with two-degree-of-freedom filters is proposed, and the corresponding filters in the internal model control are designed based on the small gain theorem. Finally, the proposed control strategy is applied to the control of the deflection angle of the electromagnetic micro-mirror to verify the proposed control method. Moreover, the non-smooth internal model control strategy is also used for comparison in the experiments.}, archive = {J_ISAT}, author = {Ruili Dong and Qingyuan Tan and Yonghong Tan and Xiaoli Song and Tianyu Wang}, doi = {10.1016/j.isatra.2025.06.003}, journal = {ISA Transactions}, month = {10}, pages = {408-421}, shortjournal = {ISA Trans.}, title = {A segmented model based internal model control scheme of electromagnetic micro-mirror systems}, volume = {165}, year = {2025}, }
• (2025). PPAC-pilot: Prescribed-performance augmented control for fixed-wing autopilots. ISAT, 165, 395-407. (www)

  @article{ , abstract = {This study introduces a Prescribed-Performance Augmented Control (PPAC) framework designed for fixed-wing Unmanned Aerial Vehicle (UAV) autopilots. The PPAC strategy aims to enhance, rather than replace, existing PID control loops in open-source autopilots. Although traditional autopilots effectively manage routine tasks in most applications, their reliance on meticulous tuning remains a limitation. To address this, PPAC leverages historical flight data, a frequently overlooked resource, to derive dynamic linearization models and control laws without requiring explicit UAV models. The PPAC framework is then integrated with the Total Energy Control System (TECS) for practical deployment in takeoff and cruising scenarios. Comprehensive numerical simulations and Hardware-in-the-Loop (HIL) tests validate the strategy by comparing baseline autopilot performance with PPAC-augmented systems. Results confirm that PPAC ensures prescribed performance bounds for altitude tracking errors across evaluated scenarios, demonstrating its effectiveness in augmenting autopilots with minimized redesign efforts.}, archive = {J_ISAT}, author = {Qiuyang Tian and Zelin Wang and Tianjiang Hu}, doi = {10.1016/j.isatra.2025.06.001}, journal = {ISA Transactions}, month = {10}, pages = {395-407}, shortjournal = {ISA Trans.}, title = {PPAC-pilot: Prescribed-performance augmented control for fixed-wing autopilots}, volume = {165}, year = {2025}, }
• (2025). Hysteresis observer enhanced integral terminal sliding mode control of piezoelectric platform for precision tracking applications. ISAT, 165, 384-394. (www)

  @article{ , abstract = {The nonlinearity of piezo-actuated positioning platforms significantly impacts their performance in high-precision applications. In this work a dynamic model of piezoelectric platform was developed firstly with the asymmetric Bouc-Wen model. Next, a terminal sliding mode observer with the super-twisting mechanism was designed to accurately estimate the state of the system. Then, a novel control strategy named Hysteresis Observer Enhanced Integral Terminal Sliding Mode Controller (HO-ITSMC) was proposed to achieve precise displacement tracking. Its stability is theoretically proved by the Lyapunov theorem. A key feature of this controller lies in its ability to drive the state of the system into zero in finite time, regardless of the initial state. Extensive experiments have thoroughly validated the effectiveness of the proposed control method, demonstrating its superior precision-tracking performance compared to traditional controllers.}, archive = {J_ISAT}, author = {Jie Chen and Lei Ni and Xuan Liao and Geng Wang and Lanqiang Zhang and Na Yao and Yijun Li and Sumeet S. Aphale}, doi = {10.1016/j.isatra.2025.06.022}, journal = {ISA Transactions}, month = {10}, pages = {384-394}, shortjournal = {ISA Trans.}, title = {Hysteresis observer enhanced integral terminal sliding mode control of piezoelectric platform for precision tracking applications}, volume = {165}, year = {2025}, }
• (2025). Adaptive robust control for ball-screw drives with flexible transmission and nonlinear friction via dynamic surface control approach. ISAT, 165, 372-383. (www)

  @article{ , abstract = {Flexible deformation and nonlinear friction in ball-screw drive systems are important factors that restrict the improvement of tracking performance. In this paper, a high-performance adaptive controller is presented for ball screw drives to suppress vibration and improve tracking accuracy. A two-inertia model with torsional vibration state is established to fit the dynamics of the drive system while the continuously differentiable LuGre model characterizes the nonlinear friction disturbance. Based on the established nonlinear model, an adaptive robust controller (ARC) is designed by using the backstepping approach to overcome the parametric uncertainties and hard-to-model dynamics. The dual-observer is employed in the controller to observe and compensate for the nonlinear friction, which improves the low-velocity tracking performance of the ball-screw drives. Meanwhile, first-order filters are introduced by dynamic surface control (DSC) technique to eliminate the “complexity explosion” problem caused by the backstepping method. The controller theoretically guarantees that all signals of the closed-loop system are bounded, and the convergence of tracking error is also ensured via Lyapunov analysis. The effectiveness of the proposed controller is verified through simulation and experimental results.}, archive = {J_ISAT}, author = {Yanliang Sheng and Guofeng Wang and Fei Wang and Decai Li and Mantang Hu}, doi = {10.1016/j.isatra.2025.05.050}, journal = {ISA Transactions}, month = {10}, pages = {372-383}, shortjournal = {ISA Trans.}, title = {Adaptive robust control for ball-screw drives with flexible transmission and nonlinear friction via dynamic surface control approach}, volume = {165}, year = {2025}, }
• (2025). A self-sensing framework for weak fault detection of planetary gearbox. ISAT, 165, 358-371. (www)

  @article{ , abstract = {Planetary gearbox fault detection has attracted wide attention due to the planetary gearbox’s key role in modern electro-mechanic equipment. However, traditional fault detection technologies still heavily rely on additional sensors. The resulting enormous cost of sensors restricts the application of those technologies. Given this situation, a self-sensing fault detection framework to explore the weak fault impulses of the planetary gearbox is presented without additional sensors. In this framework, we first capture the preliminary signals from the servo control systems. Then, the hole control model of the motor driving planetary gearbox is constructed. After this step, the feasibility of fault detection for the planetary gearbox through the motor servo control signals is investigated. With the measured servo control signals, a multi-signal assisting adaptive time synchronous averaging method is first proposed to explore fault impulses. This method first introduces a periodic enhanced Gini to select optimal parameters adaptively. Finally, experiments on a weak fault of three components in the planetary gearbox are carried out separately, certifying our framework's validation of planetary gearbox fault detection. This framework hopes to provide a novel scheme for the weak fault self-sensing of planetary gearboxes.}, archive = {J_ISAT}, author = {Dexin Chen and Ming Zhao and Shudong Ou and Sen Li and Xiaolong Han}, doi = {10.1016/j.isatra.2025.06.009}, journal = {ISA Transactions}, month = {10}, pages = {358-371}, shortjournal = {ISA Trans.}, title = {A self-sensing framework for weak fault detection of planetary gearbox}, volume = {165}, year = {2025}, }
• (2025). Nonlinear self-triggered MPC without terminal conditions for trajectory tracking. ISAT, 165, 347-357. (www)

  @article{ , abstract = {In this article, a trajectory tracking problem for a class of nonlinear discrete-time systems is investigated by a model predictive control (MPC) strategy. Compared with the standard MPC strategy, the proposed MPC strategy removes terminal conditions, including terminal penalty terms and terminal state constraints. This novel design requires fewer parameters to be determined, which leads to high practicability. Moreover, to reduce the computational burden, a self-triggered mechanism is presented by using the discrepancy in the cost function between adjacent time instants. Then, an additional compensation variable is designed for the redundancy from the self-triggered mechanism. Finally, we present a mathematical proof for the recursive feasibility of the optimization problem. The effectiveness and practicality of the proposed self-triggered MPC strategy are verified through simulation examples and experimental results on a mobile vehicle experimental platform.}, archive = {J_ISAT}, author = {Hai Zhao and Hongjiu Yang and Yuanqing Xia and Jinhui Zhang}, doi = {10.1016/j.isatra.2025.06.005}, journal = {ISA Transactions}, month = {10}, pages = {347-357}, shortjournal = {ISA Trans.}, title = {Nonlinear self-triggered MPC without terminal conditions for trajectory tracking}, volume = {165}, year = {2025}, }
• (2025). Optimal control of multi-bus DC microgrids based on distributed dual-projection-layer recurrent neural network considering bus voltage regulation. ISAT, 165, 335-346. (www)

  @article{ , abstract = {With the broad application of plug-and-play loads, it brings new challenges to conventional control issues in DC microgrids. This work addresses the joint optimization of generation costs and transmission line power losses considering bus voltage regulation. Specifically, the concept of virtual load nodes is first introduced so that loads can be implemented as plug-and-play. Through Kron Reduction, bus voltage of load nodes is indirectly controlled by DG nodes. Then, a distributed dual-projection-layer recurrent neural network (DRNN) is proposed for real-time optimal control. The coupled voltage and current are simultaneously maintained within safe bounds. By using Lyapunov synthesis, the convergence of the DRNN is demonstrated. The effectiveness of the proposed methods is evaluated by simulations in terms of plug-and-play test and comparative analysis.}, archive = {J_ISAT}, author = {Yuanyuan Zhu and Fan Yang and Guoyu Lin}, doi = {10.1016/j.isatra.2025.06.029}, journal = {ISA Transactions}, month = {10}, pages = {335-346}, shortjournal = {ISA Trans.}, title = {Optimal control of multi-bus DC microgrids based on distributed dual-projection-layer recurrent neural network considering bus voltage regulation}, volume = {165}, year = {2025}, }
• (2025). Stochastic nonlinear system modelling and parametric oscillation response characteristics of gas turbines. ISAT, 165, 320-334. (www)

  @article{ , abstract = {Gas turbines, as highly complex thermal systems, exhibit significant nonlinearity and stochastic coupling in process control. Under closed-loop automatic speed regulation, persistent parametric oscillations may arise, posing serious threats to system reliability and safety. Aiming to reveal the stochastic response characteristics of parametric oscillations in gas turbines, this paper proposes a novel framework for analyzing the evolution law of parametric oscillation and multi-source stochastic excitations based on stochastic dynamics model, which is derived from thermodynamic equations and verified by measurement data. The internal stochastic excitation is determined by information entropy, while the form of the stochastic process of the external stochastic excitation is identified through data-driven reverse identification. The PDF evolution law of parametric oscillation is studied for different excitation forms, and the bifurcation behavior and sensitivity analysis of them are carried out. Under typical operating conditions, the synergistic effect of internal and external stochastic excitations reduces parametric oscillation amplitude by approximately 31 % compared to internal excitation alone. Moreover, the originally tri-modal distribution evolves into a unimodal pattern, revealing the transition trend of parametric oscillation behavior in gas turbines. These findings offer an effective approach to analyze parametric oscillation.}, archive = {J_ISAT}, author = {Xingyun Jia and Dengji Zhou}, doi = {10.1016/j.isatra.2025.06.028}, journal = {ISA Transactions}, month = {10}, pages = {320-334}, shortjournal = {ISA Trans.}, title = {Stochastic nonlinear system modelling and parametric oscillation response characteristics of gas turbines}, volume = {165}, year = {2025}, }
• (2025). Fuzzy modal time-scheduled control and l2-gain analysis for switched nonlinear systems. ISAT, 165, 308-319. (www)

  @article{ , abstract = {This paper concerns the fuzzy modal time-scheduled control problem for switched nonlinear systems with L 2 -gain performance. Combined with the hybrid dwell time method, we propose a switched fuzzy modal time-scheduled control (FMTSC) strategy, and establish a criterion for H ∞ performance in systems comprising both unstable and stable subsystems. Meanwhile, we further develop a class of time-scheduled multiple discontinuous Lyapunov functions (TMDLFs) for the switched Takagi-Sugeno (T-S) fuzzy system with L 2 -gain property. Finally, comparative and practical examples are provided to demonstrate the validity of derived theoretical result.}, archive = {J_ISAT}, author = {Jinling Wang and Jun-Guo Lu and Jiarong Li and Qinghao Zhang and Cheng Hu}, doi = {10.1016/j.isatra.2025.06.026}, journal = {ISA Transactions}, month = {10}, pages = {308-319}, shortjournal = {ISA Trans.}, title = {Fuzzy modal time-scheduled control and l2-gain analysis for switched nonlinear systems}, volume = {165}, year = {2025}, }
• (2025). Solving a class of resource allocation problem under dynamic constraints: A predefined-time distributed optimization scheme. ISAT, 165, 295-307. (www)

  @article{ , abstract = {In this paper, a predefined-time distributed optimization algorithm is designed to solve the resource allocation problem (RAP) with dynamic constraints. This algorithm updates auxiliary variables in real time through a distributed approach and allocates resources to each node based on dynamic constraints. Its advantages include ensuring all nodes quickly converge to the optimal value within a predefined time, thereby enhancing algorithm efficiency. Moreover, the auxiliary variables exchanged between nodes do not contain any real physical information, effectively preventing privacy data leakage. In addition, the convergence of the algorithm is analyzed strictly by the Lyapunov method, which ensures the accuracy of the algorithm. Finally, application examples in smart grids and multi-UAV dynamic collaboration are provided to demonstrate the effectiveness and advantages of the algorithm in different application scenarios.}, archive = {J_ISAT}, author = {Chuxiong Su and Zhongxu Chen and Zhengyuan Zhu and Hao Dai and Jing Chang}, doi = {10.1016/j.isatra.2025.05.045}, journal = {ISA Transactions}, month = {10}, pages = {295-307}, shortjournal = {ISA Trans.}, title = {Solving a class of resource allocation problem under dynamic constraints: A predefined-time distributed optimization scheme}, volume = {165}, year = {2025}, }
• (2025). Time-optimal global path planning and collision-avoidance local path planning for USVs in traffic separation scheme-implemented coastal waters. ISAT, 165, 280-294. (www)

  @article{ , abstract = {Under multiple constraints including unmanned surface vehicle (USV) dynamics, traffic separation scheme (TSS) requirements, navigable water boundaries, and safety thresholds for collision risks, time-optimal path planning and collision-avoidance (COLAV) path planning for USVs in TSS-implemented coastal waters remain challenging. To overcome this challenge, we innovatively develop a hierarchical Gaussian-process-based nonlinear programming (GPNLP) approach for the USV time-optimal global path planning and COLAV local path planning. We model irregular static obstacles using Gaussian process regression for the first time, such that navigable waters are more sufficiently utilized for path planning. A TSS compliance assessment function is created to output violation penalties for the TSS requirements that should be satisfied as far as practicable. Accordingly, we plan the time-optimal global path and the COLAV local path hierarchically by minimizing two integral objective functions (with respect to the TSS violation penalties) subject to the multiple constraints. Simulations and simulation comparison results demonstrate that both the planned USV time-optimal global path and COLAV local path under the proposed hierarchical GPNLP approach are USV dynamics compliant and TSS compliant.}, archive = {J_ISAT}, author = {Yihan Tao and Jialu Du}, doi = {10.1016/j.isatra.2025.06.030}, journal = {ISA Transactions}, month = {10}, pages = {280-294}, shortjournal = {ISA Trans.}, title = {Time-optimal global path planning and collision-avoidance local path planning for USVs in traffic separation scheme-implemented coastal waters}, volume = {165}, year = {2025}, }
• (2025). Fixed-time switching tracking control for unmanned helicopter with multiple constraints. ISAT, 165, 268-279. (www)

  @article{ , abstract = {In this paper, a fixed-time switching tracking control scheme based on the fixed-time disturbance observer (FTDO) is proposed for a 6-DOF unmanned helicopter (UH) with multiple constraints and composite disturbances. The developed fixed-time controller guarantees that the system tracks the desired trajectory within a certain time, regardless of initial conditions. The multiple constraints include input saturation and time-varying output constraints. An improved fixed-time auxiliary system is applied to compensate for the effects of input saturation nonlinearity. By developing a novel switching boundary protection algorithm, a switching control scheme is further designed to solve the output constraints better. An improved FTDO is developed to estimate composite disturbances containing saturation function approximation errors and external disturbances. On this basis, a fixed-time switching back-stepping control method is employed for the position and attitude loops, which enables the UH to track the desired trajectory within the flight path constraints. The experimental results verify the effectiveness of the proposed scheme.}, archive = {J_ISAT}, author = {Haibo Wang and Shuang Shi and Ziyang Zhen and Ju Jiang}, doi = {10.1016/j.isatra.2025.06.017}, journal = {ISA Transactions}, month = {10}, pages = {268-279}, shortjournal = {ISA Trans.}, title = {Fixed-time switching tracking control for unmanned helicopter with multiple constraints}, volume = {165}, year = {2025}, }
• (2025). Leaderless attitude synchronization control of multiple flexible spacecraft on SO(3). ISAT, 165, 254-267. (www)

  @article{ , abstract = {This paper presents adaptive controllers for leaderless attitude synchronization of multiple flexible spacecraft on S O ( 3 ) under communication topologies containing spanning tree(s). To provide a reference attitude for each spacecraft, distributed observers in 9-dimensional Euclidean space are proposed based on a smooth mapping from Euclidean space ℝ 3 × 3 to Lie group S O ( 3 ) . Both the scenarios with arbitrary and almost zero final angular velocities are considered. Subsequently, adaptive continuous controllers on S O ( 3 ) are presented to achieve the leaderless attitude synchronization subject to external disturbance, while guaranteeing boundedness of flexible vibration. Rigorous proofs are presented to show the convergence of the proposed control methods. The effectiveness of the proposed strategies is further demonstrated by numerical simulations.}, archive = {J_ISAT}, author = {Chenlu Feng and Weicheng Jin and Ti Chen and Lifeng Wang}, doi = {10.1016/j.isatra.2025.06.031}, journal = {ISA Transactions}, month = {10}, pages = {254-267}, shortjournal = {ISA Trans.}, title = {Leaderless attitude synchronization control of multiple flexible spacecraft on SO(3)}, volume = {165}, year = {2025}, }
• (2025). Robust predefined fixed-time formation-containment control for multiple unmanned surface vehicles with input saturation. ISAT, 165, 241-253. (www)

  @article{ , abstract = {The predefined fixed-time formation-containment (PFTFC) control problem for unmanned surface vehicles (USVs) under unknown disturbances and input saturation is investigated, where the leaders form a formation configuration and the followers converge to the predefined convex hull formed by the leaders within a fixed time. To address this issue, a two-layer framework is introduced, decomposing the problem into trajectory estimation and trajectory tracking components. In the first layer, a novel fixed-time trajectory estimator is proposed for the leaders and the followers to estimate the desired trajectory. In the second layer, to eliminate the effects of unknown disturbances and input saturation, a fixed-time disturbance observer and an auxiliary system are proposed. Then combining fixed-time Lyapunov stability and integral sliding mode control, fixed-time control laws are proposed for both the leaders and the followers, respectively. Finally, a numerical simulation example is presented to illustrate the effectiveness of the method introduced in this paper.}, archive = {J_ISAT}, author = {Yong Chen and Jinlong Huang and Fuxi Niu and Xunhua Dai and Haoyue Huang}, doi = {10.1016/j.isatra.2025.06.019}, journal = {ISA Transactions}, month = {10}, pages = {241-253}, shortjournal = {ISA Trans.}, title = {Robust predefined fixed-time formation-containment control for multiple unmanned surface vehicles with input saturation}, volume = {165}, year = {2025}, }
• (2025). Constraint-following control for follow-up support systems under model deviation and time delay. ISAT, 165, 232-240. (www)

  @article{ , abstract = {Follow-up support technology can flexibly and effectively suppress machining chatter in thin-walled components. This paper proposes a servo collaborative constraint handling scheme that focuses on achieving coordinated control for the follow-up support system under multi-source coupled constraints and uncertainties. Firstly, a four-point constraint method is introduced, which simplifies the description of pose coordination through the introduction of auxiliary points, and constructs the coordinated relationship of the dual robots’ end-effectors under complex geometric constraints in an intuitive and analytical manner. Secondly, a robust controller based on constraint-following theory is designed, providing an effective means to parameterize the total uncertainty boundary structure caused by model deviation and time delay. Finally, the practical stability of the control algorithm is proven, and its effectiveness and strong robustness are validated through comparative simulations.}, archive = {J_ISAT}, author = {Fangfang Dong and Zhao Liu and Xiaomin Zhao and Jiang Han and Xiaoyong Huang}, doi = {10.1016/j.isatra.2025.06.015}, journal = {ISA Transactions}, month = {10}, pages = {232-240}, shortjournal = {ISA Trans.}, title = {Constraint-following control for follow-up support systems under model deviation and time delay}, volume = {165}, year = {2025}, }
• (2025). Stability analysis of delayed neural networks via novel delay-dependent LKF and integral inequality. ISAT, 165, 222-231. (www)

  @article{ , abstract = {The current paper is concerned with the stability analysis of delayed neural networks. In the case that the delay derivative is restricted with an upper bound only, the augmented LKFs often contain high-degree terms of the time-varying delay, resulting in the non-convex derivatives of LKFs, which can be solved by introducing extra delay-multiplied state variables to transform the non-convex delay-dependent terms into convex ones. To make fuller use of the delay-multiplied state variables and the delay-derivative-dependent information, these delay-multiplied state variables are introduced into an LKF and the integral inequality through the proper augmentation in this paper. Meanwhile, some free-matrix-based zero equations are introduced into this delay-dependent inequality to provide more freedom. By applying the augmented LKF and the novel integral inequality, a delay-dependent stability criterion of delayed neural networks with less conservatism is established, whose advantages are verified by three examples.}, archive = {J_ISAT}, author = {Fei Long and Chuan-Ke Zhang and Yanjun Shen and Qicheng Mei and Qing Chen}, doi = {10.1016/j.isatra.2025.06.027}, journal = {ISA Transactions}, month = {10}, pages = {222-231}, shortjournal = {ISA Trans.}, title = {Stability analysis of delayed neural networks via novel delay-dependent LKF and integral inequality}, volume = {165}, year = {2025}, }
• (2025). Predefined-time adaptive learning control of nonlinear strict-feedback systems via dynamic regressor extension and mixing. ISAT, 165, 209-221. (www)

  @article{ , abstract = {This paper develops a parameter identification algorithm and a novel adaptive tracking control strategy for a specific group of nonlinear strict-feedback systems incorporating the concept of predefined time under model uncertainties. A three-layer transformation-based parameter estimation method with predefined-time convergence properties is proposed to relax the strict persistent excitation condition imposed by conventional approaches. The singular terms that may occur in traditional backstepping design procedures are avoided by using a hyperbolic tangent function to design new control laws and filters. Composite learning control approach that incorporates the algorithm for parameter identification into the framework for adaptive dynamic surface control can achieve error convergence within a practical predefined time. By using Lyapunov analysis, the semi-global uniformly predefined-time boundedness for the closed-loop dynamics is demonstrated. Numerical experiments demonstrate the viability of developed control scheme.}, archive = {J_ISAT}, author = {Zhonghua Wu and Kuncheng Ma and Junkang Ni}, doi = {10.1016/j.isatra.2025.06.016}, journal = {ISA Transactions}, month = {10}, pages = {209-221}, shortjournal = {ISA Trans.}, title = {Predefined-time adaptive learning control of nonlinear strict-feedback systems via dynamic regressor extension and mixing}, volume = {165}, year = {2025}, }
• (2025). Adaptive PI nonlinear cooperative control for motor cluster. ISAT, 165, 191-208. (www)

  @article{ , abstract = {To address the effects of nonlinearities and uncertainties in the speed regulation of permanent magnet synchronous motors (PMSMs), an adaptive PI nonlinear control strategy is introduced. First, a nonlinear system model is developed using the PMSM mathematical model, and an adaptive PI nonlinear control approach is designed. Numerical simulations are conducted to demonstrate that this control method effectively tracks the system’s desired values. Then, through a group of comparative simulation experiments, the comparison effect of the designed adaptive PI nonlinear control method and the traditional PI control method is analyzed and compared. Additionally, four PMSM collaborative control system models, including the speed tracking and speed synchronization control structures, are constructed. Finally, a simulation model for a cooperative PMSM control system is developed to evaluate the system’s speed tracking capability and the synchronization between multiple motors. The results show that in the designed motor cluster cooperative control system, the PMSM motor cluster using adaptive PI nonlinear control method can achieve cooperative control in speed tracking and speed synchronization, and can maintain stable operation against nonlinear problems and unknown disturbances.}, archive = {J_ISAT}, author = {Yiting Chen and Yushen Wu and Kairui Chen and Jianhui Wang and Zian Wang}, doi = {10.1016/j.isatra.2025.05.047}, journal = {ISA Transactions}, month = {10}, pages = {191-208}, shortjournal = {ISA Trans.}, title = {Adaptive PI nonlinear cooperative control for motor cluster}, volume = {165}, year = {2025}, }
• (2025). Data-driven parallel linear controllers for reference tracking in nonlinear systems. ISAT, 165, 183-190. (www)

  @article{ , abstract = {Reference tracking control for nonlinear systems presents significant challenges, particularly when system models are unavailable and real-time computation is required. We present a purely data-driven approach for reference tracking control, referred to as parallel linear controllers (PLIC), which leverages an architecture composed of two linear controllers operating concurrently in distinct dimensional spaces. These controllers are employed for inverse control and mismatch error compensation, respectively. The first controller utilizes the Koopman operator for lifting the system to a high dimension and solves a quadratic program that facilitates constraint handling. The second controller, which works in the original state space, employs the direct data-driven virtual reference tuning approach with some appropriate modifications. The closed-loop properties of the proposed PLIC are analyzed, and the efficacy of the proposed method is exemplified through benchmark simulation.}, archive = {J_ISAT}, author = {Yao Shi and José M. Maestre and Lei Xie and Hongye Su}, doi = {10.1016/j.isatra.2025.05.048}, journal = {ISA Transactions}, month = {10}, pages = {183-190}, shortjournal = {ISA Trans.}, title = {Data-driven parallel linear controllers for reference tracking in nonlinear systems}, volume = {165}, year = {2025}, }
• (2025). Real-time random reference tracking nonlinear model predictive control: A case study on wind turbines. ISAT, 165, 170-182. (www)

  @article{ , abstract = {Recently, a research effort to extend nonlinear model predictive control methods from setpoint stabilization to reference tracking has been felt increasingly. On the other hand, uncertainty in the reference signal and the requirement for its dynamic forecasting in applications such as wind turbine control motivate the need for robust tracking nonlinear model predictive control approaches more and more. Therefore, this study proposes a random reference tracking nonlinear model predictive control with dynamic forecasting of stochastic references. Convergence to a robust invariant set is guaranteed by an additional constraint limiting the previous step’s tracking stage cost function. The proposed predictive approach is implemented using a parallel Newton-type method to make it more efficient and applicable. The proposed approach for wind turbine control is designed considering the random wind speed reference. Simulations are performed for extreme and fatigue load scenarios. The results show that the proposed controller performs more robustly than the nominal nonlinear model predictive control approach, performing better in optimal power extraction and reducing aerodynamic loads.}, archive = {J_ISAT}, author = {Mohammad Soleymani and Nooshin Bigdeli and Mehdi Rahmani}, doi = {10.1016/j.isatra.2025.06.018}, journal = {ISA Transactions}, month = {10}, pages = {170-182}, shortjournal = {ISA Trans.}, title = {Real-time random reference tracking nonlinear model predictive control: A case study on wind turbines}, volume = {165}, year = {2025}, }
• (2025). A novel MPC-based cascaded control for multi-area smart grids: Tackling renewable energy and EV integration challenges. ISAT, 165, 143-169. (www)

  @article{ , abstract = {This paper presents an advanced cascaded control scheme for load frequency regulation in multi-area power systems incorporating renewable energy sources (RES) and electric vehicles (EVs). The proposed design (Model predictive control cascaded with one plus proportional-integral control cascaded with tilt control in parallel with one plus fractional-order integral derivative controller (MPC-((1+PI)-(T+(1+I λ D μ )))) combines predictive, tilt, and fractional-order dynamics to improve adaptability and robustness under uncertainties. Controller parameters are tuned using the Lyrebird Optimization Algorithm (LOA), ensuring fast convergence and effective global search. Simulation results under varying operational conditions, including nonlinearity effects such as Generation Rate Constraints (GRC), Governor Dead Band (GDB), and Communication Time Delays (CTD), confirm the controller’s superiority. It achieves a 96.4 % ITAE reduction, 98.6 % undershoot mitigation, and a settling time of just 5.8 s outperforming existing benchmark strategies (GOA: PDf+(0.75+PI), CBOA: PI-PD, JSA: PI, and ARA: 1+PID).}, archive = {J_ISAT}, author = {Muhammad S. Tolba and Muhammad Majid Gulzar and Ali Arishi and Mohamed Soliman and Ali Faisal Murtaza}, doi = {10.1016/j.isatra.2025.06.024}, journal = {ISA Transactions}, month = {10}, pages = {143-169}, shortjournal = {ISA Trans.}, title = {A novel MPC-based cascaded control for multi-area smart grids: Tackling renewable energy and EV integration challenges}, volume = {165}, year = {2025}, }
• (2025). Frequency handling by reinforcement of predictive 2DoF-MPC and state observer LADRC for smart power system. ISAT, 165, 128-142. (www)

  @article{ , abstract = {The preservation of stability is essential for the efficient and reliable functioning of the electrical transmission system. Frequency oscillations are prevalent in interconnected power systems (IPS) and may lead to instability; therefore, it is crucial to monitor and examine them meticulously. Effective frequency management is essential for regulating frequency output in an interconnected smart power system (ISPS) that includes renewable energy sources (RESs), redox flow batteries (RFBs) and static synchronous series compensators (SSSCs). In view of the challenge presented, this research introduces an efficient control architecture that utilizes a 2 degree of freedom-based model predictive controller (2DoF-MPC) to enhance system performance. Additionally, it integrates a linear active disturbance rejection control (LADRC) to employ a state observer alongside the evolving frequency management. The convergence of the predictive and state observer frameworks results in a robust 2DoF-MPC-LADRC to manage frequency disturbances and uncertainties in the power system. The suggested technique is thoroughly validated across several parameters for ISPS, instilling confidence in its capacity to attain minimal frequency variation in multiple scenarios. The performance of the proposed controller design shows that the frequency performance in area 1 and area 2 settles in 5.085 sec and 3.965 sec, when the load changes by 3 %, and it settles in 4.655 sec and 4.050 sec, respectively, when the load changes by 5 %.}, archive = {J_ISAT}, author = {Muhammad Majid Gulzar}, doi = {10.1016/j.isatra.2025.05.046}, journal = {ISA Transactions}, month = {10}, pages = {128-142}, shortjournal = {ISA Trans.}, title = {Frequency handling by reinforcement of predictive 2DoF-MPC and state observer LADRC for smart power system}, volume = {165}, year = {2025}, }
• (2025). Fractional-order sliding mode coordinated controller using super-twisting disturbance observer for an NSSS with predefined-time stability. ISAT, 165, 111-127. (www)

  @article{ , abstract = {This paper develops a fractional-order sliding mode coordinated control (FOSMCC) strategy incorporating dual super-twisting disturbance observers (STDOs) to enhance the control performance, stability, and reliability of the nuclear steam supply system (NSSS) under complex, time-varying operating conditions and compound disturbances. The FOSMCC strategy synthesizes the fractional-order control and predefined-time theory with sliding mode control, augmented by the disturbance feedforward compensation loop driven by dual STDOs. Such control framework provides enhanced control performance guarantees, including fast transient response, high steady-state precision, and reinforced disturbance rejection. Furthermore, by employing Lyapunov’s direct approach, it is theoretically demonstrated that the entire NSSS, under the developed coordinated strategy, achieves superior predefined-time stability. Finally, comprehensive numerical validation and comparative studies reveal that the developed FOSMCC strategy with STDOs significantly outperforms both the latest fractional-order fixed-time sliding mode controller (FOFTSMC) and the practically adopted coordinated controller (PACC), exhibiting better transient/steady-state control response and stronger robustness against disturbances. Simulation results validate that, in the presence of compound disturbances, the proposed FOSMCC strategy reduces the integral absolute control error of nuclear power and water level by 89.37 % and 87.67 %, respectively, compared to FOFTSMC, and by 99.97 % and 99.99 %, respectively, compared to PACC.}, archive = {J_ISAT}, author = {Jiuwu Hui}, doi = {10.1016/j.isatra.2025.06.032}, journal = {ISA Transactions}, month = {10}, pages = {111-127}, shortjournal = {ISA Trans.}, title = {Fractional-order sliding mode coordinated controller using super-twisting disturbance observer for an NSSS with predefined-time stability}, volume = {165}, year = {2025}, }
• (2025). Fixed-time observer-based saturated nonsingular terminal sliding mode controller design for an over-actuated ROV with time-varying saturation limits. ISAT, 165, 98-110. (www)

  @article{ , abstract = {Input saturation is critical in over-actuated systems when multiple degrees of freedom (DOF) with different levels of disturbance rejection are controlled simultaneously by a set of actuators. The current study introduces a novel saturated non-singular terminal sliding mode controller to address input saturation for an over-actuated remotely-operated vehicle (ROV). The proposed controller consists of a tuning algorithm to ensure that the control commands do not violate the time-varying saturation limits of each DOF. In addition, a fixed-time extended-state observer is designed to estimate the vehicle’s velocity along with the lumped unknown dynamics of the system. The observer is also employed as a tool to maintain the orientation of the ROV in extreme environmental conditions. The stability analysis shows that all system’s states, except for yaw angle, are globally finite-time stable and the yaw angle is globally asymptotically stable. Several sets of simulations are carried out and the results demonstrate the superiority of the proposed controller in terms of positioning accuracy, saturation compensation and transient behaviour under different environmental conditions.}, archive = {J_ISAT}, author = {Alireza Hosseinnajad and Navid Mohajer}, doi = {10.1016/j.isatra.2025.06.025}, journal = {ISA Transactions}, month = {10}, pages = {98-110}, shortjournal = {ISA Trans.}, title = {Fixed-time observer-based saturated nonsingular terminal sliding mode controller design for an over-actuated ROV with time-varying saturation limits}, volume = {165}, year = {2025}, }
• (2025). A nonsingular fast terminal sliding mode control scheme for robust trajectory tracking of the underactuated EvoBot modular mobile robot in the vertical plane. ISAT, 165, 83-97. (www)

  @article{ , abstract = {A robust control strategy for the vertical plane motion of an underactuated EvoBot mobile robot, modeled as a serial double inverted pendulum, is presented in this study. The control objectives include controlling both the directly actuated generalized coordinates and the non-actuated generalized coordinate, which is controllable through dynamic coupling with the actuated coordinates. In this regard, the system's dynamic equations are derived using the Lagrangian approach. A new nonlinear and nonsingular sliding manifold is introduced, based on which a nonsingular fast terminal sliding mode control scheme is proposed for the trajectory tracking control of the robot. This approach addresses the challenges posed by underactuation, system nonlinearities, instability, parameter uncertainties, and external disturbances. Through Lyapunov stability analysis, it is proven that finite-time asymptotic convergence of the tracking error to zero is ensured when the uncertainty upper bound is known, and convergence to a residual set is achieved when the upper bound is unavailable. The theoretical guarantees provided by the proposed control scheme are further validated through comprehensive MATLAB simulations, where its effectiveness is demonstrated under both low- and high-frequency disturbances as well as parameter uncertainties.}, archive = {J_ISAT}, author = {H. Jokar and S. Amini Serajgah}, doi = {10.1016/j.isatra.2025.06.013}, journal = {ISA Transactions}, month = {10}, pages = {83-97}, shortjournal = {ISA Trans.}, title = {A nonsingular fast terminal sliding mode control scheme for robust trajectory tracking of the underactuated EvoBot modular mobile robot in the vertical plane}, volume = {165}, year = {2025}, }
• (2025). Finite-time event-triggered sliding mode control for fuzzy singular systems under cyber-attacks. ISAT, 165, 72-82. (www)

  @article{ , abstract = {This paper investigates a novel secure control scheme for a particular class of Takagi–Sugeno (TS) fuzzy singular systems susceptible to deception attacks. During these attacks, adversaries can randomly introduce erroneous data into the output and control signals. The proposed strategy addresses the impact of attacks and disturbances using an observer-based sliding mode control (SMC) approach. Moreover, an event-triggering protocol is implemented to manage network resources efficiently. Furthermore, by employing the stochastic Lyapunov theory and the finite-time analysis method, sufficient conditions are established to ensure the finite-time boundedness of the resulting closed-loop system throughout both the reaching and sliding motion phases. To mitigate the attack’s effects and improve the system’s performance, the Secretary Bird Optimization Algorithm (SBOA) with the linear matrix inequality (LMI) is explored as a new approach for designing the optimal gains of controllers and observers. Finally, a simulation study based on a disc rolling on a surface is performed to showcase the efficacy and resilience of the proposed control scheme.}, archive = {J_ISAT}, author = {Mourad Kchaou and Rabeh Abassi and Houssem Jerbi}, doi = {10.1016/j.isatra.2025.06.012}, journal = {ISA Transactions}, month = {10}, pages = {72-82}, shortjournal = {ISA Trans.}, title = {Finite-time event-triggered sliding mode control for fuzzy singular systems under cyber-attacks}, volume = {165}, year = {2025}, }
• (2025). Memory-based dynamic event-triggered secure control of active suspension systems against deception attacks. ISAT, 165, 64-71. (www)

  @article{ , abstract = {This study deals with the issue of a memory-based dynamic event-triggered control strategy for active quarter-vehicle suspension systems (QVSSs). The main objective is to design an effective event-triggered mechanism (ETM) that ensures suspension performance while reducing network resource usage, even under deception attacks. To this end, an innovative memory-based dynamic ETM is proposed to coordinate sensor data transmission efficiently in the presence of such attacks. The proposed transmission scheme integrates historical release information, which helps suppress false triggering by utilizing averaged data. Additionally, the proposed ETM dynamically updates triggering conditions over time, facilitating dynamic scheduling of network data transmission. Sufficient conditions are derived to guarantee satisfactory performance of the QVSS under the proposed control strategy. A numerical example is provided to validate the effectiveness of the approach.}, archive = {J_ISAT}, author = {Wangrui Cheng and Tingting Yin and Zhou Gu}, doi = {10.1016/j.isatra.2025.06.007}, journal = {ISA Transactions}, month = {10}, pages = {64-71}, shortjournal = {ISA Trans.}, title = {Memory-based dynamic event-triggered secure control of active suspension systems against deception attacks}, volume = {165}, year = {2025}, }
• (2025). Time-varying formation control for heterogeneous multi-agent systems in the presence of actuator faults and deception attacks. ISAT, 165, 54-63. (www)

  @article{ , abstract = {This paper explores the control of time-varying formations in a class of heterogeneous multi-agent systems. The key innovation lies in the simultaneous consideration of hybrid actuator faults and deception attacks. To achieve the control objective, a novel distributed double-layer control scheme, comprising a network layer and a physical layer, is proposed. In the network layer, a distributed observer with secure output feedback control is developed to mitigate severe deception attacks, ensuring that the mean square observer error remains within an acceptable range. In the physical layer, fault compensators are designed to address both additive and multiplicative faults. As a result, the followers achieve time-varying formation control, and closed-loop stability analysis is conducted using the Lyapunov method. Finally, to verify the validity of the theoretical findings, numerical simulations are subsequently conducted.}, archive = {J_ISAT}, author = {Shicheng Cao and Yanhui Yin and Wenyu Li and Zhongxin Liu and Zengqiang Chen}, doi = {10.1016/j.isatra.2025.06.004}, journal = {ISA Transactions}, month = {10}, pages = {54-63}, shortjournal = {ISA Trans.}, title = {Time-varying formation control for heterogeneous multi-agent systems in the presence of actuator faults and deception attacks}, volume = {165}, year = {2025}, }
• (2025). Event-triggered consensus of multi-agent systems with uncertain control gain via distributed fuzzy logic observer. ISAT, 165, 40-53. (www)

  @article{ , abstract = {An event-triggered adaptive backstepping control methodology is proposed to achieve leader-following consensus of uncertain nonlinear high-order multi-agent systems with unknown control gains. In light of the partial observability limitation, adaptive distributed observers are employed to estimate the unobservable states of the leader, whereas local state observers are utilized to reconstruct the states of the followers. By integrating fuzzy logic systems, the unknown nonlinear dynamics are modeled, guaranteeing reliable state prediction in complex and partially observable scenarios. Moreover, the novel relative threshold event-triggered scheme is designed to reduce the frequency of data interactions while ensuring that the tracking error approaches near-zero. Eventually, the effectiveness and superiority of the devised controller are clearly demonstrated through comprehensive simulation results.}, archive = {J_ISAT}, author = {Konghao Xie and Xiujuan Zhao and Shiming Chen and Zheng Zhang and Yuanshi Zheng}, doi = {10.1016/j.isatra.2025.06.020}, journal = {ISA Transactions}, month = {10}, pages = {40-53}, shortjournal = {ISA Trans.}, title = {Event-triggered consensus of multi-agent systems with uncertain control gain via distributed fuzzy logic observer}, volume = {165}, year = {2025}, }
• (2025). Consensus seeking in large-scale multi-agent systems with homogeneous connections by incorporating two-hop neighbor states. ISAT, 165, 27-39. (www)

  @article{ , abstract = {The development of multi-agent consensus raises the importance of network topology. As the number of agents increases, multi-agent systems (MAS) in a large-scale and high-density topology demand higher resources, which consequently degrades efficiency of consensus. Existing approaches that consider only direct point-to-point neighbors may overlook potential topological information, further hindering consensus performance. To achieve fast consensus in large-scale and high-density topologies, a framework named Homogeneous Connections Based on Agents State Fusions MAS (HCASFMAS) is proposed. The framework extracts broader topology information of consensus degree by fusing states of two-hop neighbors. Leveraging homogeneous idea, agents establish homogeneous connections with neighbors that exhibit a higher consensus degree, ultimately accelerating the consensus process while preserving connectivity. First, a neighbor selection strategy based on consensus degree of agent state fusion is introduced to construct candidate neighbors, aiming to reduce redundant connections. Second, an adaptive consensus algorithm is formulated to flexibly adapt to the distribution of neighbors. Finally, a candidate constraints set is established to accelerate consensus by expanding the scope of constraints while preserving connectivity. In this study, connectivity and convergence of the system are theoretically analyzed from a geometric perspective. Simulation experiments are conducted to compare the proposed method with existing approaches under different densities and topologies. Simulation results demonstrate the superiority of this method in achieving fast convergence, particularly in large-scale and high-density scenarios.}, archive = {J_ISAT}, author = {Guangqiang Xie and Chaohao Shen and Yang Li and Yanda Feng and Fengyang Qiu}, doi = {10.1016/j.isatra.2025.06.002}, journal = {ISA Transactions}, month = {10}, pages = {27-39}, shortjournal = {ISA Trans.}, title = {Consensus seeking in large-scale multi-agent systems with homogeneous connections by incorporating two-hop neighbor states}, volume = {165}, year = {2025}, }
• (2025). Dynamic formation tracking and fault-tolerant control of multi-agent systems based on distance and topology reconstruction methods. ISAT, 165, 15-26. (www)

  @article{ , abstract = {This paper introduces a distributed approach for dynamic formation tracking and formation fault-tolerant control within the port-Hamiltonian energy framework for multi-agent system (MAS) affected by Coulomb friction. The coupling relationships between agents are equivalently modeled as virtual springs, which simulate the interaction forces between agents to reflect the relative positions and motion states of the agents. A distance-based distributed control scheme is designed, to ensure that the formation composed of multiple agents can continuously adjust the direction and size of the formation while achieving target tracking. Additionally, considering the possibility of communication failure due to agent motion faults, a fault-tolerant algorithm based on topological reconstruction is proposed to reconstruct the formation topology after faults. The feasibility of this control method is verified through numerical simulations.}, archive = {J_ISAT}, author = {Lu Liu and Yu Cui and Yonghua Wang and Jinglong Wen and Shuaishuai Kong and Yuhang Ma and Dan Liu and Peng Shi and Chenyang Xue}, doi = {10.1016/j.isatra.2025.06.006}, journal = {ISA Transactions}, month = {10}, pages = {15-26}, shortjournal = {ISA Trans.}, title = {Dynamic formation tracking and fault-tolerant control of multi-agent systems based on distance and topology reconstruction methods}, volume = {165}, year = {2025}, }
• (2025). Fast actuator fault-tolerant control for a class of nonlinear sampled-data systems via deterministic learning. ISAT, 165, 1-14. (www)

  @article{ , abstract = {In this paper, we investigate the fast fault-tolerant control (FTC) problem based on deterministic learning approach (DLA) for a class of nonlinear sampled-data systems with actuator faults, which consist of two stages: incipient faults with small magnitudes and faults with larger magnitudes. First, a learning controller and a learning identifier are constructed. Based on DLA and the exponential stability of a class of linear time-varying (LTV) discrete-time systems, the control knowledge and the diagnosis knowledge of the actuator faults are obtained. Second, a set of controllers and a set of diagnosis estimators are constructed based on the learnt control and diagnosis knowledge. When an incipient actuator fault occurs, fast fault detection and isolation (FDI) can be achieved using the diagnosis estimators. Then, the pattern-based FTC scheme is implemented to improve the control performance. When the small fault grows to a larger one, the rapid FDI and FTC are implemented again, providing fast responses to the occurred larger fault. The advantages of the proposed method are that: (i) a simple adaptive learning controller with the filtering technique is designed, in which the exponential convergence of the tracking error and parameter estimation errors can be achieved simultaneously; (ii) the sensitivity to small actuator faults is enhanced, and the fast FTC to larger actuator faults is achieved by utilizing the learnt knowledge. Simulation results are also included to illustrate the effectiveness of these schemes.}, archive = {J_ISAT}, author = {Yu Zeng and Tianrui Chen and Fukai Zhang and Cong Wang}, doi = {10.1016/j.isatra.2025.05.049}, journal = {ISA Transactions}, month = {10}, pages = {1-14}, shortjournal = {ISA Trans.}, title = {Fast actuator fault-tolerant control for a class of nonlinear sampled-data systems via deterministic learning}, volume = {165}, year = {2025}, }